AI-Based Learning Platforms: A Systematic Review of Evaluation Metrics for Accessibility, Interactivity and Adaptability through the Lens of Universal Design for Learning

AI-Based Learning Platforms: A Systematic Review of Evaluation Metrics for Accessibility, Interactivity and Adaptability through the Lens of Universal Design for Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Systematic Review AI-Based Learning Platforms: A Systematic Review of Evaluation Metrics for Accessibility, Interactivity and Adaptability through the Lens of Universal Design for Learning Frank Alexander Parra Sánchez, Juan Pablo Rivera Barrera, Henry Omar Sarmiento Maldonado, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8801285/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Artificial Intelligence is reshaping the design, evaluation, and personalization of digital learning environments, enabling adaptive and data-driven pedagogies that respond to diverse learner needs. In parallel, the Universal Design for Learning (UDL) framework has become central to inclusive education, offering principles to ensure accessibility, engagement, and multiple means of representation. Despite this convergence, systematic analyses that evaluate AI-based learning platforms through UDL remain scarce. Existing reviews on AI in education have primarily focused on algorithmic efficiency, adaptive architectures, or technological innovation. However, they lack an analytical framework that connects AI-based learning technologies with UDL principles, largely because they do not articulate the dimensions needed to operationalize these principles into evaluative criteria. As a result, most AI-driven platforms are assessed in terms of technical performance rather than pedagogical inclusiveness, usability, accessibility compliance, or learner engagement. To address this gap, this study conducts a systematic review of evaluation metrics applied to AI-based learning platforms, following the PRISMA methodology and analyzing peer-reviewed studies published from 2019 onward. Using UDL as a conceptual and analytical scaffold, the review structures its synthesis around three operational dimensions derived from the framework: accessibility (representation), interactivity (action and expression), and adaptability (engagement and motivation). Building on this analytical approach, the purpose of the review is twofold: first, to map the state of the art in evaluating AI-driven learning platforms through both normative and algorithmic metrics; and second, to propose an integrative model that links international standards with user-experience indicators and adaptive performance measures. In doing so, the study contributes a structured evaluative perspective that bridges technical methodologies and pedagogical frameworks for inclusion, advancing the development of more equitable, transparent, and accessible AI-based learning systems. Artificial Intelligence Universal Design for Learning Inclusive Learning Environments Accessibility Interactivity Evaluation Metrics Figures Figure 1 Figure 2 Figure 3 Figure 4 1 Introduction Artificial Intelligence (AI) has increasingly shaped contemporary educational technologies, redefining the ways in which digital platforms are designed, evaluated, and personalized. Recent advances in learning analytics, adaptive algorithms, and intelligent feedback systems have enabled the development of environments capable of responding dynamically to learners’ profiles, behaviors, and performance patterns (Kenneth et al., 2024 ; Sajja et al., 2025 ). Within this landscape, the UDL framework has consolidated its role as one of the most influential approaches for promoting inclusion in technology-mediated education. By proposing multiple means of representation, action and expression, and engagement, UDL provides a pedagogical foundation to design equitable environments that accommodate diverse needs from the outset (CAST, 2018 ). Although AI and UDL have evolved largely in parallel, their intersection presents an emerging opportunity: integrating data-driven adaptivity with principles of accessibility and inclusive design. This convergence is especially relevant in a context where digital learning platforms increasingly mediate instruction, assessment, and interaction. However, despite rapid progress in AI-enhanced learning systems, the literature still lacks systematic analyses that connect these technologies with UDL-based evaluative criteria. Existing reviews have largely concentrated on algorithmic performance, adaptive learning architectures, or technological innovation (such as learning analytics, recommender systems, or intelligent tutoring systems) without examining how these systems align with pedagogical principles of accessibility, interactivity, and adaptability (Khosravi et al., 2023 ; Tan et al., 2025 ; Tebourbi et al., 2025 ). This disjointed focus has produced a methodological gap: most AI-driven educational platforms are evaluated through computational indicators (accuracy, RMSE, precision/recall), while human-centered dimensions such as usability, accessibility compliance, satisfaction, and learner engagement remain secondary (ISO & IEC, 2016; Montes et al., 2024 ). In other words, although AI systems increasingly influence learning processes, the frameworks used to assess them rarely consider whether they support inclusive pedagogical practices or embody the flexibility promoted by UDL. To address this limitation, the present study conducts a systematic review of evaluation metrics and assessment frameworks applied to AI-based learning platforms, following the PRISMA guidelines and examining peer-reviewed studies published from 2019 onward. UDL is used as the conceptual and analytical scaffold, enabling the organization of findings around three operational dimensions derived from the framework: accessibility (representation), interactivity (action and expression), and adaptability (engagement and motivation). This conceptual structure allows the review to analyze not only how platforms are technically evaluated, but also how current metrics reflect (or fail to reflect) the pedagogical commitments of inclusive design (Barbu et al., 2025 ; Ingavélez-Guerra et al., 2023 ; Mehmood, 2025 ). Accordingly, this review seeks to map the state of the art in evaluating AI-driven learning platforms and to propose an integrative model that brings together international standards (e.g., ISO, WCAG), user-experience indicators, and adaptive performance measures. By bridging technical methodologies and pedagogical frameworks for inclusion, the study contributes to building evaluative criteria capable of guiding the development of more equitable, transparent, and accessible AI-based educational systems. Specifically, this review addresses the following research questions: How has recent literature evaluated AI-based learning platforms in terms of accessibility, interactivity, and adaptability? What metrics, instruments, or reference frameworks are used to measure these dimensions? How can these findings inform the design of inclusive learning environments based on UDL principles? The article is organized as follows: Section 2 outlines the theoretical framework connecting UDL principles with international quality metrics for educational software. Section 3 describes the PRISMA-based methodology used for study identification, selection, and analysis. Section 4 presents the results structured around the three operational dimensions of UDL. Finally, Section 5 discusses the implications of these findings for evaluating and designing inclusive AI-driven learning platforms. 2 Related works Recent research in educational technology reveals a fragmented yet rapidly evolving convergence between AI, accessibility standards, and inclusive pedagogy. Existing systematic reviews and empirical studies cluster around four dominant lines. The first concerns digital accessibility and quality assurance, emphasizing persistent inconsistencies between international frameworks (such as WCAG 2.1 and ISO/IEC 25022) and their actual implementation in educational platforms. Automated auditing tools often produce divergent outcomes for the same resources, particularly in STEM environments, where accessibility barriers in navigation, contrast, and non-text content remain unresolved (Ismailova & Inal, 2022 ; Zapata & Acosta, 2019 ). These limitations are compounded by educators’ limited awareness of accessibility and UDL principles. A second body of literature focuses on AI-driven personalized learning and adaptive assessment, where platforms increasingly employ machine learning algorithms (such as Fuzzy C-Means, Random Forest, or Deep Knowledge Tracing) to infer cognitive states and adapt content dynamically (Lu, 2024 ). Adaptive Learning Platforms (ALPs) and intelligent tutoring systems demonstrate significant gains in knowledge retention, yet often prioritize prediction accuracy over inclusivity and transparency (Parveen et al.). The third line addresses multimodal and gesture-based interfaces for inclusive interaction. Speech, text, and gesture recognition systems have enhanced accessibility for diverse learners, though continuous sign-language recognition remains a major technical challenge requiring robust datasets and cross-linguistic validation (Barkovska et al., 2025 ; Bercaru & Popescu, 2024 ). Finally, a growing strand examines accessibility metadata in open and distance learning (ODL). Despite the existence of structured standards such as IMS AfA V3.0 and ISO/IEC 24751, their integration into MOOCs and e-learning systems remains limited (Ingavélez-Guerra et al., 2023 ), while learner satisfaction and persistence continue to depend heavily on perceived accessibility and usability (Obeid et al., 2024 ; Punitham et al., 2025 ). Taken together, these studies illustrate remarkable technological progress but also expose a persistent methodological gap: the absence of an integrative framework that connects accessibility, interactivity, and adaptability, the operational core of UDL. Although prior reviews have examined aspects of AI-enhanced education, none has systematically analyzed the evaluation metrics linking these three dimensions in the context of AI-based learning platforms. Addressing this gap, the present study proposes a unified approach that bridges normative quality standards (WCAG, ISO/IEC 25022), algorithmic performance indicators, and user-experience analytics to advance a coherent model for evaluating inclusive, AI-driven learning environments. 3 Theoretical framework This section presents the conceptual foundations that guide the analysis of AI-based learning platforms from an inclusive and evaluative perspective. The framework draws on three complementary components: (a) UDL as the pedagogical foundation for inclusion; (b) AI as the technological infrastructure enabling personalization and automation; and (c) international quality standards (ISO/IEC and WCAG) as reference models for assessing software usability, accessibility, and effectiveness. Together, these perspectives provide an integrated lens to understand how intelligent learning environments can be evaluated not only for their technical performance but also for their capacity to promote equitable, accessible, and adaptive learning. 3.1 Fundations of UDL The UDL framework is one of the most influential approaches to inclusive education in contemporary pedagogy. Developed by (CAST, 2018 ), UDL is based on the premise that learner diversity is inherent to any educational process, and therefore instructional design must anticipate variability rather than respond to it retroactively. UDL proposes three core principles: Multiple means of representation (the what of learning), which promote offering information in varied perceptual and linguistic formats; Multiple means of action and expression (the how of learning), which broaden the ways students interact with content and demonstrate understanding; and Multiple means of engagement (the why of learning), which sustain motivation, interest, and persistence. These principles align with broader global commitments to equitable and quality education, including Sustainable Development Goal 4 (UNESCO, 2019 ). In digital environments, UDL reframes diversity as a design criterion rather than as an instructional constraint. This perspective has been increasingly used to guide the development of digital materials, pedagogical strategies, and formative assessment practices (Parody et al., 2022 ; A. Saborío-Taylor & G. Rojas-Ramírez, 2024 ). When digital learning platforms incorporate adaptive technologies, UDL becomes especially pertinent: its principles can be operationalized through multimodal interfaces, personalized feedback, and algorithmic adjustments that support learner variability. 3.2 AI and UDL: Toward Inclusive Intelligent Learning Environments AI-based learning platforms have introduced new possibilities for personalization, including recommendation systems, adaptive difficulty adjustment, conversational tutors, and predictive analytics. These systems can analyze learner interactions and adjust instructional pathways accordingly. However, technological sophistication alone does not guarantee inclusiveness; systems may optimize performance metrics without supporting equitable participation or diverse learning needs (Sajja et al., 2025 ). The integration of AI and UDL thus represents an opportunity to redesign digital learning with both personalization and inclusion in mind. While UDL provides the pedagogical foundation for removing barriers through flexible representation, expression, and engagement, AI provides the technical mechanisms to enact these principles at scale and in real time (M. Saborío-Taylor & J. Rojas-Ramírez, 2024 ). This convergence transforms UDL from a design philosophy into an evaluative lens: the behavior of the system (its feedback, adaptivity, and multimodal mediation) can be assessed in terms of how well it reflects UDL’s commitments to accessibility, interactivity, and learner-centered engagement (El-Sabagh & Hamed, 2021 ). 3.3 From UDL Principles to the Operational Dimensions of Accessibility, Interactivity, and Adaptability Although UDL is formulated as a pedagogical framework, its principles can be translated into operational dimensions that enable empirical evaluation in digital environments. This translation is consistent with the sociocognitive view of learning presented by Gee, who argues that comprehension is grounded in perceptual, embodied, and contextual processes rather than in the decoding of linguistic symbols alone. According to Gee, understanding emerges through “perceptual simulations that prepare agents for situated action” (Barsalou, 1999 ; Gee, 2001 ) and through “experiences (perceptions, feelings, actions, and interactions) stored as dynamic images tied to perception” (Gee, 2001 ). These ideas align directly with UDL’s emphasis on offering multiple means of representation, reinforcing accessibility as the dimension that ensures diverse perceptual and linguistic pathways for understanding information. The principle of multiple means of action and expression corresponds to the dimension of interactivity, which captures the extent to which learners can act upon and with the system. Gee highlights that learning is inseparable from speaking, listening, and interacting, with the material and social world, underscoring that meaning is constructed through embodied interaction rather than passive reception. Moreover, he notes that language scaffolds the performance of action in the world (Gee, 2001 ), supporting the relevance of feedback-rich, dialogic, and manipulable interfaces-core characteristics of interactivity in digital learning platforms. Finally, the principle of engagement in UDL maps onto adaptability, understood as the system’s capacity to support motivation, persistence, and autonomy through personalized learning conditions. Gee states that meaning is always customized to our actual contexts, purposes, values, and intended courses of action, emphasizing that learners interpret information according to their goals and prior experiences. He explains this adaptive process through the metaphor of running our tapes (Gee, 2001 ), where learners draw on previous embodied experiences to engage with new situations. This notion parallels the adaptive mechanisms in AI-based platforms that adjust content, support, and difficulty levels in response to learner profiles and evolving engagement. Taken together, Gee’s sociocognitive perspective supports the operationalization of UDL principles into the dimensions of accessibility, interactivity, and adaptability by showing that learning is inherently perceptual, interactive, and context-dependent. These dimensions therefore provide a coherent and empirically grounded structure for evaluating whether AI-based learning environments reflect the inclusive intentions of UDL in their design and behavior. 3.4 International Standards (ISO/IEC and WCAG) as Complementary Evaluation Frameworks Building on the UDL-derived dimensions described above, intelligent learning environments must also adhere to established technical quality standards that guide usability, accessibility, reliability, and “quality in use.” The ISO/IEC 9126 and ISO/IEC 25000 SQuaRE series offer widely recognized frameworks for software evaluation, including effectiveness, efficiency, satisfaction, and freedom from risk (Botella et al., 2004 ; ISO & IEC, 2016). These standards have been applied extensively in the evaluation of educational software and digital interfaces, providing measurable criteria for verifying user performance in real-world contexts (ISO, 2019 ). Similarly, the Web Content Accessibility Guidelines (WCAG 2.1) define criteria for perceptibility, operability, understandability, and robustness in online environments, establishing minimum accessibility levels for inclusive digital content (W3C 2018). Research on accessibility technologies for deaf learners has demonstrated the value of aligning digital resources with WCAG to ensure equitable interaction, especially when sign language or multimodal interfaces are involved (Saunders et al. 2019; Huenerfauth et al. 2008). When combined with UDL, ISO and WCAG frameworks offer complementary evaluative structures: Accessibility aligns with WCAG and ISO effectiveness/efficiency criteria. Interactivity corresponds to ISO measurements of productivity, error recovery, and system responsiveness. Adaptability links to ISO constructs such as reliability, flexibility, and satisfaction, which capture system responsiveness to user variability (Ifenthaler & Gibson, 2020 ; ISO, 2019 ). Thus, ISO and WCAG standards reinforce UDL’s pedagogical commitments with verifiable technical metrics. 3.5 An Integrative Framework for the Evaluation of Intelligent Learning Environments Bringing these components together, UDL principles, AI capabilities, and international evaluation standards, creates a coherent analytical model for understanding inclusive intelligent learning environments. Prior work has emphasized the need for frameworks that bridge pedagogical intent with technological implementation in AI-driven systems (El-Sabagh & Hamed, 2021 ; VanLehn, 2006 ). In this study, UDL defines what inclusive systems must guarantee (accessibility, meaningful interaction, and adaptive engagement), while AI provides the mechanisms to enact these commitments through multimodal mediation, personalization algorithms, and intelligent feedback (Adomavicius & Tuzhilin, 2005 ; M. Saborío-Taylor & J. Rojas-Ramírez, 2024 ). ISO/IEC and WCAG standards, in turn, offer measurable and internationally validated criteria for assessing usability, accessibility, and quality in use. This integrated framework underpins the analytical model applied in the systematic review. It guides the identification, classification, and interpretation of evaluation metrics reported in AI-based learning platforms and supports the broader goal of advancing inclusive, transparent, and pedagogically aligned digital learning environments (L. Herrera Nieves et al., 2025 ; Ingavélez-Guerra et al., 2023 ; Mehmood, 2025 ). To summarize these connections, Table 1 links the three UDL-derived dimensions with their corresponding ISO/IEC and WCAG quality metrics. This synthesis provides a concise framework for evaluating inclusiveness, usability, and adaptability in AI-based learning environments. Table 1 Correspondence between UDL principles, operational dimensions, and ISO/WCAG quality metrics Dimension Main purpose In UDL In ISO/WCAG (software quality) Predominant variable type Accessibility Ensuring that all users can access the learning environment and understand the information. Representation principle. ISO 9241 − 210 → perceptibility, usability; WCAG 2.1 → perceivable, operable, understandable, robust; ISO 25022 → effectiveness, efficiency, context coverage. Structural (entry conditions). Interactivity Enabling the user to act on the system and receive meaningful, timely feedback. Action and expression principle. ISO 9126-4 / ISO 25010 → productivity, response time, feedback effectiveness, error recovery. Processual (use dynamics). Adaptability Adjusting the experience to the user (profile, progress, motivation, or cognitive/affective state). Engagement principle. ISO 25022 / ISO 9126 → satisfaction, reliability, flexibility, freedom from risk. Personalizing (intelligent adjustment). 4 Methodology This study followed the Preferred Reporting Items for Systematic Reviews and Meta- Analyses (PRISMA 2020) guidelines (Page et al., 2021 ) to ensure methodological transparency and reproducibility. The review process comprised four sequential stages: (1) definition of the research question and eligibility criteria, (2) systematic search in indexed databases, (3) screening and coding of retrieved studies, and (4) qualitative and quantitative synthesis of evidence. 4.1 Study design The review adopted an exploratory–descriptive design aimed at identifying and analyzing evaluation metrics applied to AI-based learning platforms through the principles of UDL. The protocol was registered internally and validated by two domain experts before execution, guaranteeing alignment between the research questions and the coding categories derived from the theoretical framework (accessibility, interactivity, and adaptability). 4.2 Search Strategy and Information Sources The search was conducted between 2024 and 2025 across six major databases: Scopus, IEEE Xplore, Web of Science, ScienceDirect, SpringerLink, and ACM Digital Library. Boolean search equations were tailored to each database to maximize precision and recall. Representative strings included: ("AI-based learning platform*" OR "intelligent tutoring system*" OR "adaptive learning system*") AND ("evaluation metric*" OR "assessment framework*" OR "usability" OR "learning analytic*") AND ("accessibility" OR "interactivity" OR "adaptability") AND ("Universal Design for Learning" OR "UDL") Only peer-reviewed articles published between 2019 and 2025 were considered, reflecting the post-pandemic acceleration of AI adoption in education. Additional records were retrieved through backward snowballing from the reference lists of key studies. 4.3 Selection and data-coding process All retrieved documents were exported to Zotero for deduplication and then screened independently by two reviewers in three rounds: title/abstract, full text, and methodological appraisal. Eligibility followed the inclusion/exclusion criteria summarized in Table 2. Disagreements were resolved through discussion and consensus. To guarantee reliability, intercoder agreement was calculated using Cohen’s κ, yielding κ = 0.84, which indicates substantial agreement according to (Landis & Koch, 1977). This statistical control reinforces the consistency of decisions across reviewers and minimizes subjectivity in classification. Table 2 Correspondence between UDL principles, operational dimensions, and ISO/WCAG quality metrics Inclusion Criteria (IC) Exclusion Criteria (EC) IC1. Studies describing or evaluating educational platforms assisted by artificial intelligence. EC1. Studies describing platforms without the use of AI or related techniques. IC2. Research addressing at least one of the three components: accessibility, interactivity, or adaptability. EC2. Studies focused solely on pedagogical content without considering technical aspects of the platform. IC3. Publications between 2018 and 2025, peer-reviewed, in English or Spanish. EC3. Publications prior to 2018 or non-scientific articles (e.g., posters, blogs, etc.). IC4. Articles describing technical functionalities or AI tools implemented in educational environments. EC4. Purely theoretical studies without application or validation of the proposed technology. IC5. Open-access articles. EC5. Articles with access limitations. 4.4 Data extraction and coding scheme A structured coding matrix was developed based on the analytical dimensions of UDL and on the ISO/IEC 25022 quality-in-use indicators. Each article was coded across 20 variables, including publication year, country, educational level, AI techniques, evaluation metrics, accessibility compliance, and type of data used. The coding process was supported by Atlas.ti 23, which facilitated the identification of co-occurring themes and the visualization of conceptual clusters. Memos and code families were employed to track the relationships between AI methods and UDL principles. The final corpus comprised 24 studies meeting all inclusion criteria. Each record was assigned a unique identifier and cross-checked against the database source to prevent duplication. 4.5 Analysis and synthesis of information The synthesis combined descriptive statistics (frequency and co-occurrence distributions) with qualitative thematic analysis of conceptual and methodological trends. Quantitative data were processed in Python 3.12 using pandas and matplotlib libraries to compute frequency tables and figures. Qualitative interpretation focused on how evaluation frameworks operationalized the UDL dimensions of accessibility, interactivity, and adaptability. Figure 1 illustrates the PRISMA flow diagram summarizing identification, screening, eligibility, and inclusion stages. 4.6 Use of generative artificial intelligence tools During the preparation of this manuscript, the authors made limited use of generative artificial intelligence tools exclusively to support language editing and stylistic refinement. These tools were employed solely to improve clarity, coherence, and grammatical accuracy of the text, without contributing to the study design, data analysis, interpretation of results, or generation of substantive scientific content. All content was critically reviewed and edited by the authors, who assume full responsibility for the accuracy, originality, and integrity of the published work. 5 Results 5.1 General Characteristics The analysis of the 24 selected articles made it possible to identify, from the perspective of UDL principles, how recent literature addresses the evaluation of AI-based learning platforms, focusing on the criteria of accessibility, interactivity, and adaptability. Figure 2 shows the proportion of studies that address each of these dimensions. Adaptability is the most frequently explored (42%), followed by interactivity (33%) and accessibility (25%). This pattern confirms that research attention is concentrated on the personalization and algorithmic adjustment of the learning experience, while universal accessibility components remain less explored. Regarding the educational level of application (see Fig. 3 ), most of the studies are situated in higher education contexts (13 articles), followed by mixed-level studies (4 articles) and, to a lesser extent, experiences in primary and secondary education (2 articles). Seven studies do not specify the educational level of implementation. This trend highlights a significant gap in research targeting early school levels, where UDL principles could have a particularly relevant impact on inclusion and motivation for learning. Figure 4 presents the combinations of features analyzed. Most studies do not focus on a single dimension but rather integrate multiple dimensions simultaneously. The most frequent combination is adaptability + interactivity (11 articles), followed by the integration of all three dimensions (8 articles). In contrast, only two studies focus exclusively on a single category (one on accessibility and one on adaptability). This finding reflects a clear trend toward multidimensional approaches, in which personalization and interaction converge, while accessibility continues to play a complementary rather than a structural role. Regarding the evaluation metrics, the analyzed studies can be grouped into three categories: international standards and guidelines (WCAG 2.1, ISO 25022, ISO 9241); usability and technology acceptance models (SUS, TAM, UTAUT, UEQ); and algorithmic and performance metrics (efficiency, effectiveness, error rate, engagement, feedback accuracy, precision/recall, RMSE). Table 3 presents this classification, establishing the relationship between the evaluation metrics, their purposes, their correspondence with UDL dimensions, and the authors who support them. Table 3 Classification of metrics and standards used in the reviewed studies Reference group Examples of metrics or models Evaluation focus / UDL mapping (with sources) International standards and guidelines WCAG 2.1; ISO/IEC 25022:2016; ISO 9241–210 Establish technical criteria for accessibility, usability, and software quality (effectiveness, efficiency, user satisfaction, flexibility). This aligns mainly with accessibility (representation) and adaptability (efficiency, flexibility). Reported in (Ara et al., 2024 ; L. Herrera Nieves et al., 2025 ; Ingavélez-Guerra et al., 2023 ; ISO & IEC, 2016). Usability and technology acceptance models SUS (System Usability Scale); TAM (Technology Acceptance Model); UTAUT (Unified Theory of Acceptance and Use of Technology); UEQ (User Experience Questionnaire) Evaluate user perception, ease of use, satisfaction, acceptance, and learning experience in AI-mediated environments. These instruments map to interactivity (action and expression) and adaptability (engagement, autonomy, satisfaction). (Zhang et al., 2024 ). Algorithmic and performance metrics Efficiency, effectiveness, error rate, engagement , feedback accuracy , precision/recall , RMSE Measure AI algorithm performance, personalization effectiveness, and the system’s predictive capability in real time. These metrics correspond to adaptability (dynamic adjustment and adaptive learning) and interactivity (personalized feedback and participation analysis). Reported in (Demartini et al., 2024 ; Isaeva et al., 2025 ; Mehmood, 2025 ). The convergence between international standards (such as ISO/IEC 25022 and WCAG 2.1), technology acceptance models (TAM, UTAUT, SUS), and algorithmic performance metrics (efficiency, precision, error rate, and engagement) reveals a clear trend toward the triangulation of technical, cognitive, and pedagogical approaches in the evaluation of AI-based educational platforms. In the reviewed literature, these metrics do not operate in isolation; rather, they are articulated in a complementary manner to assess quality, usability, and learning impact in alignment with the UDL principles. This landscape suggests that evaluation processes are shifting from a merely functional perspective of software toward an integrated assessment of the educational experience, where technical performance, student engagement, and personalized learning are analyzed as interdependent dimensions. Nevertheless, the results also highlight gaps in the systematization of the applied criteria: while ISO standards and WCAG guidelines provide standardized indicators of accessibility and quality in use, machine learning metrics and user perception scales tend to operate in a fragmented way, without an explicit alignment with UDL principles. The following section deepens the analysis by categories, examining how the reviewed studies address the three critical dimensions that operationalize UDL principles (accessibility, interactivity, and adaptability) identifying the most representative metrics, techniques, and approaches that support their empirical evaluation. 5.2 Analysis by Category 5.2.1 Accessibility Accessibility represents the dimension most closely aligned with the UDL principle of multiple means of representation, ensuring that all users (regardless of their sensory, cognitive, or contextual conditions) can effectively perceive, understand, and navigate digital environments. In AI-assisted platforms, this principle is operationalized through inclusive design strategies, international quality standards, and assistive technologies aimed at eliminating or reducing barriers to information access. Among the 24 studies analyzed, only 25% address accessibility as a central focus, confirming its lower prevalence compared to adaptability and interactivity. However, the studies that do incorporate it approach accessibility through three complementary perspectives: (1) compliance with international technical standards, (2) integration of assistive and multimodal accessibility tools, and (3) perceptual evaluation of inclusive usability. The first group of studies associates accessibility with the structural foundation of digital inclusion, including verification of compliance with international standards (WCAG 2.1, ISO 9241, ISO/IEC 25022) and the incorporation of governance, ethics, and privacy principles as part of inclusive design. (Maksymov et al., 2025 ), for example, address accessibility through a quantitative assessment of technical compliance. Their proposed methodology includes explicit metrics on the availability of accessibility features and support programs (technical, financial, or technological), translating regulatory compliance into measurable equity indicators. In a more normative dimension, (Madanchian & Taherdoost, 2025 ; Mehmood, 2025 ) broaden the concept toward ethics and governance, emphasizing that digital equity also depends on data privacy, user security, and the objectivity of AI-based systems. These perspectives combine technical compliance metrics with trust and transparency indicators, aligning accessibility with its role as a digital right. The second group of studies conceptualizes accessibility as a dynamic condition in which AI acts as a mediator to provide multiple and sensorially adaptive representations. (Memari & Taheri, 2024 ) exemplify this approach with their adaptive sign language teaching model for Iranian Sign Language, using metrics such as word weight, repetition, speed, and user score to adjust content difficulty based on user performance, applying fuzzy logic to personalize the experience. Similarly, (Sheejamol et al., 2025 ) work with neurodivergent populations (ASD, ADHD, dyslexia) and demonstrate that adaptive, multisensory gamification improves engagement and knowledge retention, reinforcing the link between accessibility and cognitive adaptability. Table 4 expand the concept into Extended Reality (ER) environments, integrating voice, text, and gesture modalities to create immersive and accessible educational experiences. Although their metrics focus on task completion rates, their overall objective is to transform accessibility into an immersive and participatory experience, consistent with UDL’s representation principle. Likewise, (Kenneth et al., 2024 ) emphasize the role of teachers in multimodal inclusion, underscoring that accessibility requires not only adaptive technologies but also pedagogical competencies to mediate their implementation. Their study highlights the need for teacher training in inclusive AI tools (e.g., OneNote, Socratic, Quizlet), positioning accessibility as a formative (not merely technical) process. The third approach combines user experience assessment with the institutional support needed to ensure equitable access and the sustainability of inclusive design. (Maksymov et al., 2025 ), for instance, consider navigation speed and number of clicks as efficiency indicators, incorporating institutional support programs as an additional criterion and extending the evaluation to the organizational dimension. Other studies, such as those by (Sajja et al., 2025 ) and (Rajabi, 2025 ), frame usability within Scientific and UX Design, measuring affective states such as stress, curiosity, and confusion to adapt personalized interventions, while integrating indicators of robust technical infrastructure and system acceptance as key elements of user satisfaction. Finally, (Morgado et al., 2025 ) and (Isaeva et al., 2025 ) highlight the importance of teacher and institutional support, emphasizing continuous teacher training and material diversity as essential conditions to sustain accessibility and meaningful learning. Based on the above, the findings indicate that accessibility in AI-based platforms should not be understood merely as the availability of technical resources, but rather as an integrated construct that encompasses regulatory compliance, multimodal inclusion, and inclusive usability. In this way, accessibility ceases to be an isolated starting point and becomes a cross-cutting indicator of inclusive quality, which supports the subsequent dimensions of the model: interactivity and adaptability. This perspective is complemented by the identification of a set of principles or requirements that the authors, either explicitly or implicitly, highlight in their studies and that should be considered when designing accessible learning platforms, along with the guiding questions to achieve this. Table 4 summarizes this information Table 4 Principles, requirements, and guiding questions for accessible platforms Author and year Principles or requirements (explicit or implicit) Guiding questions for accessible design (Maksymov et al., 2025 ) Accessibility and inclusivity as a universal evaluation criterion; adherence to WCAG and international standards; evaluation of mobile accessibility; institutional support for inclusion. Does the platform meet accessibility requirements (WCAG) for users with visual, auditory, motor, or cognitive disabilities? Does the platform provide text descriptions for videos and images? (Barbu et al., 2025 ) Compliance with accessibility standards in ER; WCAG 2.2 and W3C ER Accessibility User Requirements; generation of AI-driven content adaptation for students with special educational needs. Does the design framework adhere to WCAG 2.2 and W3C ER Accessibility User Requirements to ensure inclusion? Does the system dynamically adapt learning environments to each learner’s needs and progress? (Kenneth et al., 2024 ) UDL to dynamically adjust content, interaction, and engagement; legal compliance (Section 508 and WCAG); need for teacher training to balance technologies and promote inclusion. Are the accessibility criteria stipulated by WCAG and Section 508 being met through AI-driven personalization? Are educators being trained to integrate AI-based technologies and balance automation with pedagogical oversight? (Sajja et al., 2025 ) Cognitive accessibility for neurodivergent learners (ASD, ADHD, dyslexia); mitigation of sensory overload; multimodal strategies (text, visual, audio, haptic); integration with UDL. Is the UI/UX design minimizing cognitive overload in neurodivergent learners? Does the system adapt to individual cognitive profiles instead of forcing a one-size-fits-all model? (Memari & Taheri, 2024 ) Continuous and multimodal adaptation for sign languages (e.g., Iranian Sign Language); continual learning to expand vocabulary without catastrophic forgetting; fuzzy logic for personalized teaching parameters (playback speed, repetition, scoring). Does the system adapt modality parameters (speed, repetition) to each learner’s performance capacity? How does the AI system manage vocabulary recognition accuracy while continuously expanding its lexicon? (Rajabi, 2025 ) Robust technical infrastructure and high-quality UX as success factors for adaptation across diverse populations; multimodal prompts. Is the technological framework robust and user-friendly enough to support personalized and adaptive learning for diverse student populations? (Mehmood, 2025 ) Ethical governance and inclusivity; ensuring that AI systems are fair and understandable. Are the AI systems fair, understandable, and inclusive? Do governance principles prevent malpractices and protect the rights of all students? (Morgado et al., 2025 ) Digital inclusion and equity of access regardless of socioeconomic or geographic context. Have concrete strategies been implemented to guarantee digital inclusion and platform accessibility for all students? 5.2.2 Interactivity (UDL Principle of Action and Expression) Interactivity constitutes one of the most critical operational pillars for understanding how AI–based platforms foster active student participation in learning. In alignment with the second principle of UDL, providing multiple means of action and expression, this category examines the capacity of intelligent environments to enable bidirectional communication, stimulate critical thinking, and provide meaningful feedback. The reviewed studies show that interactivity, when mediated by AI algorithms, is structured around three key subdimensions: (1) cognitive interactivity, which fosters reasoning and self-regulation; (2) social interactivity, which enables collaboration and peer learning; and (3) algorithmic interactivity, which enhances the user experience through automated feedback, response pattern analysis, and task personalization. The first subdimension focuses on the accuracy and personalization of feedback that the system provides to the learner, functioning as an intelligent tutor that adjusts guidance and difficulty in real time according to the user’s performance, cognitive states, and affective states. In this line, (Memari & Taheri, 2024 ) propose an adaptive teaching model for Iranian Sign Language (ISL), applying fuzzy logic and continual learning to adjust parameters such as word weight, repetition, and sign speed. This architecture enables effective and profile-sensitive interaction, achieving a significant improvement in perceived adaptability (Cohen’s d = 0.63). Similarly, (Tan et al., 2025 ) integrate the Revised Bloom’s Taxonomy (RBT) into intelligent tutoring systems, automatically adjusting the difficulty level of tasks and questions to sustain students’ cognitive progression, while (Khosravi et al., 2023 ) develop the RiPPLE system, which uses AI to provide personalized hints and explanations in peer learning activities (learnersourcing). This approach turns feedback into a formative process of self-regulation and collective knowledge construction. (Tebourbi et al., 2025 ) introduce the concept of dynamic scaffolding through the AIA-PAL framework, based on Multi-Agent Systems (MAS) and Large Language Models (LLMs). This model adjusts support levels as learners progress, ensuring a balance between challenge and guidance. Complementarily, (Kenneth et al., 2024 ) examine AI-Powered Personalized Learning (AI-PPL) and its capacity to keep learners within their zone of proximal development through real-time feedback in tools such as Socratic. Other studies, such as (Wangdi & Shimray, 2025 ) and (Slepankova et al., 2025 ), emphasize the importance of immediate feedback in sustaining engagement and com- prehension. However, they caution that superficial or shallow feedback can limit cognitive depth. (Halkiopoulos & Gkintoni, 2024 ) advocate for intelligent feedback mechanisms based on cognitive neuropsychology principles, which adjust content to the user’s thinking style and emotional state. The second subdimension explores the ability of AI to facilitate collaboration and peer communication, promoting social learning and community building in virtual environments. In this regard, (Sajja et al., 2025 ) examines emotional interaction and its relationship with engagement through VirtualTA, a tool that uses GPT-4 and natural language processing to identify affective states such as stress, curiosity, confusion, and agitation. This emotional profiling enables personalized interventions that sustain collective motivation. In the domain of collaborative learning, (Khosravi et al., 2023 ) report that the RiPPLE system increases trust in peer assessment by reducing bias through identity anonymization and reinforcing the perception of fairness. Only 2% of users expressed disagreement with the scores received, demonstrating the model’s reliability. Likewise, (Barbu et al., 2025 ) extend social interaction to immersive ER environments, where virtual agents or NPCs adapt their responses and tone to the user’s communicative style, fostering social skill development in inclusive contexts. (Isaeva et al., 2025 ) highlight the role of simulations, virtual labs, and educational games in active learning, while (Sheejamol et al., 2025 ) show that collaborative gamification strengthens communication and negotiation among neurodivergent students, promoting a participatory and inclusive environment. The final subdimension focuses on technical fluency and UX-factors that determine the perceived quality and sustainability of interaction. (Maksymov et al., 2025 ) identify usability and interface design as central criteria in the evaluation of educational platforms, prioritizing simplicity, minimalism, and reducing the number of steps required to complete a task. (Han et al., 2024 ) link System Quality (SQ) to continued use intention, showing that the fluidity of interactive functions and logical consistency are essential to maintain user motivation. From a technical perspective, (Tebourbi et al., 2025 ) measure agent response time and workflow synchronization, demonstrating that delays or poorly timed transitions reduce the perceived naturalness of interaction. (L. B. Herrera Nieves et al., 2025 ), in their evaluation of Moodle, emphasize the importance of clear organization, a comprehensible interface, and functional aesthetics, confirming that perceived usability is an essential component of interactivity. (Madahana et al., 2022 ) argue that usability should be intuitive and accessible to non-expert users, enabling teachers to integrate AI tools without requiring advanced technical knowledge. Finally, (Rajabi, 2025 ) and (Demartini et al., 2024 ) highlight the relationship between robust technical infrastructure and learning effectiveness, demonstrating that environments with high interactivity and low cognitive load foster user satisfaction and retention. Overall, the studies indicate that interactivity in AI-driven platforms has evolved from simple user–system exchange toward cognitive, social, and emotionally adaptive communication, aligned with UDL’s principle of action and expression. The three identified requirements ((1) cognitive dialogue and adaptive feedback, (2) AI-mediated social interactivity, and (3) interactive usability and system responsiveness) constitute an articulated framework that integrates intelligent feedback, meaningful collaboration, and technical fluency. Interactivity, when mediated by AI, not only optimizes learning efficiency but also redefines learner agency as active participation in a human–algorithmic interaction network. In this context, feedback quality, computational empathy, and user experience become critical indicators of equity and participation, positioning interactivity as an essential vector of inclusion and engagement in contemporary educational environments. In the analyzed studies, interactivity emerges as a balancing component between automation and human agency. The most successful platforms do not replace student participation but rather amplify it through intelligent mediations that integrate dialogue, adaptability, and collaboration. The identified metrics make it possible to quantify the quality of human–machine interaction, while triangulation with ISO standards (particularly 25022 and 9241) provides a technical structure to measure user efficiency and satisfaction. Based on these ideas, these approaches demonstrate that interactivity constitutes the connection point between the pedagogical principles of UDL and the algorithmic capabilities of AI. Table 5 synthesizes the principles or requirements identified around interactivity and the design questions formulated by the authors to guide the development of AI-based learning environments from pedagogical, communicative, and technological perspectives. Table 5 Principles, requirements, and guiding questions for the design of interactive platforms Author and Year Principles or Requirements (explicit or implicit) Guiding Questions for Interactive Design (Memari & Taheri, 2024 ) Effective Human–Agent Interaction (HAI); adaptive teaching programs based on user performance; fuzzy logic to adjust system parameters (speed, repetition). Does the adaptive teaching system interact effectively with diverse users and achieve the desired adaptability in training sessions? How can learning parameters (playback speed, repetition) be optimized based on user performance (speed, accuracy, score)? (Wangdi & Shimray, 2025 ) Need for detailed explanatory feedback in Self-Access Language Learning (SALL) environments beyond multiple-choice interactions. How can SALL platforms be improved to provide detailed feedback and explanations for incorrect answers, reducing reliance on MCQs? How can limitations in learners’ self-access experience be addressed through adaptive interactivity? (Demartini et al., 2024 ) Valuable feedback and continuous content adaptation aligned with classroom dynamics; Intelligent Decision Support Systems (IDSS) for corrective actions (e.g., flipped classrooms, tutoring). What methods and tools are used to collect and process student data to inform interaction? What corrective actions can an IDSS provide to teachers or administrators? (Halkiopoulos & Gkintoni, 2024 ) Continuous, adaptive feedback mechanisms; use of engagement metrics to adjust learning experience in real time. How can attention and perception principles be used to create AI systems that personalize and enhance learning interaction? How can cognitive neuroscience help guide AI models toward more effective learning environments? (Han et al., 2024 ) SQ, including recognition accuracy (e.g., in drawing questions) and logical difficulty progression; emphasis on perceived ease of use (PEU) and feedback. How can the adaptive engine be optimized to improve question logic and recognition accuracy, increasing users’ Continuous Intention (CI) to use the system? Which perceived external control factors (teacher/technical support) are essential to sustain CI? As a synthesis of this section, interactivity in AI-powered platforms emerges as a key dimension for materializing the UDL principle of action and expression. The reviewed studies agree that the quality of interaction depends not only on the number of exchanges but also on their cognitive and affective depth. The integration of intelligent agents, multimodal feedback, and emotional analysis enables the development of more autonomous, dialogic, and personalized learning experiences. However, ethical and pedagogical challenges remain regarding how to balance automation with human support. 5.3 Adaptability (UDL Principle: Engagement and Involvement) Adaptability emerges as the most prominent category in the reviewed literature and functions as the central interface between AI and Universal UDL. In alignment with UDL’s third principle (providing multiple means of engagement and sustaining learner motivation) this dimension refers to the capacity of AI-driven systems to dynamically adjust content, strategies, and learning pathways in response to each learner’s characteristics, interests, and performance. Across studies, adaptability is not conceptualized as a static software property but rather as a continuous process of cognitive, emotional, and contextual adjustment requiring both algorithmic precision and intentional pedagogical design. Collectively, the literature identifies four core requirements that operationalize this dimension: dynamic adjustment of content and difficulty levels, learner modeling and predictive personalization, sustained motivation and engagement, and equity and diversity in adaptive processes. Regarding the first requirement, the evidence shows that adaptability functions through real-time modifications of content, learning pathways, and task complexity, supported by intelligent tutoring strategies and dynamic scaffolding. For example, (Tan et al., 2025 ) establish the conceptual basis of adaptive sequencing and assessment by measuring learning gains between pre- and post-tests. Their model applies the Revised Bloom’s Taxonomy (RBT) to adjust task difficulty and cognitive progression, enabling activities to evolve from lower-order skills to complex creative tasks. In an applied context, (Memari & Taheri, 2024 ) developed an adaptive platform for teaching Iranian Sign Language (ISL), using fuzzy logic and Elastic Weight Consolidation (EWC) to personalize the user experience. Their metrics (word weight, repetition frequency, and execution speed) allowed precise measurement of adaptive fit and interaction fluency, yielding a significant perceived adaptability effect (Cohen’s d = 0.63). Likewise, (Hssina & Erritali, 2019 ) conceptualize adaptation as an optimization problem solvable through genetic algorithms that search for optimal learning pathways based on the distance between the learner’s current profile and target competencies. (Tebourbi et al., 2025 ) expand this approach through a MAS implementing dynamic scaffolding, progressively adjusting support levels in response to real-time interactions. Finally, (Kenneth et al., 2024 ) emphasize that AI-PPL tools should deliver immediate adaptive feedback to sustain progression within each learner’s zone of proximal development. The second requirement positions learner modeling as the technical core of adaptability, integrating cognitive, behavioral, and emotional data to predict performance and inform pedagogical strategies. For instance, (Demartini et al., 2024 ) propose a closed-loop regulatory environment that dynamically updates learner profiles based on interaction data, enriched with indicators of socio-emotional competencies. This allows pedagogical decisions to be informed by a more holistic understanding of learning trajectories. (Halkiopoulos & Gkintoni, 2024 ) argue that personalization should be guided by cognitive neuropsychology, incorporating cognitive styles and profiles into Adaptive Assessment (AA) systems. Empirical evidence from (Khosravi et al., 2023 ) supports this approach: their RiPPLE system leverages learner sourcing data to feed predictive models, demonstrating superior performance compared to systems relying solely on traditional assessments. These results highlight increased accuracy and relevance of personalized recommendations. The third requirement emphasizes the emotional dimension of adaptability, understood as the system’s ability to sustain learner motivation, self-efficacy, and positive attitudes through dynamic interventions. Drawing on the TAM3 model, (Han et al., 2024 ) identify Perceived Usefulness and Perceived Enjoyment as the most influential factors in CI to use adaptive environments, with Computer Self-Efficacy mediating the relationship between ease of use and persistence. Affectively, (Sajja et al., 2025 ) developed a real-time emotion analysis system using GPT-4 to detect stress, curiosity, confusion, and agitation in learners’ language. This continuous monitoring allows adaptive adjustment of both content and feedback types to sustain cognitive engagement. (Mehmood, 2025 ) extends this discussion to mindset formation, arguing that adaptive systems can strengthen not only academic performance but also positive learning attitudes and emotional resilience. These findings are reinforced by (Memari & Taheri, 2024 ), who report significant increases in learner motivation and willingness to interact with adaptive systems. Finally, the fourth requirement suggests that adaptability must go beyond technical personalization and performance to ensure equity, inclusion, and algorithmic ethics, avoiding biases that perpetuate inequalities. (Sheejamol et al., 2025 ) argue that traditional one-size-fits-all e-learning models are ineffective for neurodivergent learners. They propose multimodal personalization and adaptive gamification models responsive to diverse cognitive styles and sensory needs. (Maksymov et al., 2025 ) reinforce this perspective by including adaptability to individual learning pace, multilingual localization, and institutional flexibility as measurable quality indicators. At an organizational level, (Madanchian & Taherdoost, 2025 ) emphasize that adaptability must be scalable and sustainable, ensuring effective AI implementation across different class sizes and educational contexts. (Kenneth et al., 2024 ) and (L. B. Herrera Nieves et al., 2025 ) add that inclusive adaptation is achieved when both pedagogical and technical design incorporate UDL principles, ensuring accessibility for all learners. As in the previous sections, Table 6 presents the principles or requirements and the guiding questions based on the analyzed authors’ works, which should be considered when designing user-adaptive platforms. Table 6 Principles, requirements, and guiding questions for designing adaptive platforms Author & Year Principles or Requirements (explicit or implicit) Guiding questions raised by the authors for adaptive design ( Hssina & Erritali, 2019 ) Algorithmic adaptability through genetic algorithms; personalized learning paths according to profile and objectives. How can learning paths evolve according to each student’s progress and learning style? ( Han et al., 2024 ) Affective and cognitive adaptation through multimodal emotion recognition. How can difficulty and content be adjusted based on the learner’s emotional state and attention? ( Rajabi, 2025 ) Adaptability based on machine learning and ISO standards; mobile and cognitive personalization. How can personalization enhance the experience without compromising fairness and transparency? ( Tebourbi et al., 2025 ) Adaptability through cooperative agents (AIA-PAL); dynamic learning paths with human validation. What balance should exist between AI-driven automation and teacher supervision? ( Memari & Taheri, 2024 ) Linguistic adaptability through continual learning (EWC); personalization of learning pace and vocabulary. How can cumulative learning be maintained without forgetting prior knowledge? ( Sajja et al., 2025 ) Cognitive and emotional adaptability; personalized paths based on learner progress and affective states. How can AI trace itineraries that reflect students’ cognitive and emotional development? ( Wangdi & Shimray, 2025 ) Hierarchical adaptability (CEFR levels); automatic progression according to performance. What mechanisms allow the platform to evaluate and adjust the student’s level in real time? ( Sheejamol et al., 2025 ) Gamified adaptability (PAGE model); adjustment of difficulty level and rewards according to learning style. How can personalization, motivation, and performance be balanced for neurodivergent students? ( Barbu et al., 2025 ) Sensory and cognitive adaptability in ER environments; 3D environment personalization according to performance and preference. How should immersive environments respond to diverse learning paces and styles? Based on the analysis of the reviewed articles, adaptability emerges as the most developed dimension within AI-assisted educational platforms. The authors address this principle through algorithmic personalization, cognitive and affective adaptation, and continual learning, proposing systems that dynamically adjust content, difficulty, and pace according to student performance. These strategies aim to operationalize the UDL engagement principle by enabling flexible learning paths that are sensitive to individual differences. In summary, the literature shows that adaptability not only enhances personalization but also raises critical challenges related to equity, teacher supervision, and the sustainability of learning. While advances in machine learning and student modeling allow for immediate responses to individual needs, the challenge remains to ensure that these adaptations do not result in fragmented or decontextualized experiences, but rather strengthen autonomy, motivation, and the continuity of knowledge. 5.4 Cross-sectional Analysis of the Three Dimensions Based on the 24 articles included in the review, recent literature increasingly addresses the evaluation of AI-assisted learning platforms from a systemic perspective that integrates three interdependent dimensions of the UDL: accessibility (representation), interactivity (action and expression), and adaptability (engagement). Rather than being analyzed separately, these dimensions function as linked elements of a pedagogical quality cycle. Coverage analysis indicates that adaptability receives the most attention (42%), followed by interactivity (33%) and, to a lesser extent, accessibility (25%). Most studies focus on higher education contexts (12 studies), with fewer in primary/secondary education (2 studies), four mixed-scope proposals, and seven unspecified, highlighting a research gap in school stages where UDL could have a more direct impact. The convergence across dimensions is evident at multiple levels. Accessibility has evolved beyond mere regulatory compliance (WCAG 2.1, ISO 9241/ISO 25022) to integrate with interactivity and adaptability through AI-mediated supports, such as automatic captioning, sign language interpreters, multimodal interfaces, and ER accessibility features. These interventions ensure that accessible representation translates into meaningful action and active participation, promoting equity in learning. Interactivity bridges access and personalization, turning available content into effective cognitive and social experiences. Adaptive feedback, dynamic tutoring, and AI-enabled social mediation (learner sourcing, scaffolding, and decision support ecosystems) enhance learning relevance and engagement. Interactive usability and technical fluency ensure that access opportunities are effectively utilized, while affective reading allows real-time adjustment according to students’ emotional states. Adaptability serves as the integrative core, linking technical personalization with pedagogical inclusion. Student modeling, adaptive sequencing, continuous learning, and real-time scaffolding optimize content difficulty and pacing while coordinating accessibility and interactivity. These adaptations address neurodiversity, multilingual needs, scalability, and the AI–teacher balance, ensuring personalized experiences that are equitable and pedagogically meaningful. Evaluation metrics converge across the three dimensions into a comprehensive framework: Standards and regulations: WCAG 2.1, ISO 9241, ISO/IEC 25022, ensuring effectiveness, efficiency, context coverage, and usability. Usability and acceptance models: SUS, UEQ, TAM/TAM3, UTAUT, capturing user perceptions of ease-of-use, satisfaction, and engagement. Algorithmic and educational performance metrics: feedback accuracy, response time, precision/recall, RMSE, semantic analysis of interactions, pre/post Learning Gains, Bloom/RBT mapping, error and dropout rates, participation/collaboration measures, and real-time affective indicators. In ER environments and neurodivergent contexts, additional requirements address cognitive load, multisensory responses, and data governance, demonstrating the convergence of technical, pedagogical, and ethical criteria. In summary, the evidence indicates that AI-based learning platform design should operate systemically, leveraging the interplay of accessibility, interactivity, and adaptability through integrated metrics. This approach guides inclusive, adaptive, and student-centered designs, ensuring learning is meaningful, equitable, and motivating. Design Guidelines from UDL Empirical evidence suggests a concrete roadmap: Accessibility as an entry condition: combine normative auditing (WCAG; ISO 9241/25022) with functional testing in diverse populations (e.g., deaf users), verifying effectiveness/efficiency, context coverage, and understandability (SUS/UEQ). Include linguistic accessibility (interpreters/sign avatars, captions) and cognitive accessibility (minimalist UI, load reduction). Interactivity as pedagogical mediation: ensure explanatory and timely feedback (beyond correct/incorrect) with quality traceability; affective reading for personalized interventions; meaningful collaboration (peers/NPCs); and technical fluency (response times, coordination, low errors) to support student agency. Adaptability as ethical and sustainable adjustment: operate with robust learner models (cognitive–behavioral–affective), adaptive sequencing/evaluation (Bloom/RBT), dynamic scaffolding (MAS/LLMs), and bias controls (multilingual, devices, subgroups) with teacher supervision and governance policies. Based on the analysis of the reviewed literature, it is possible to establish a roadmap to guide the design of AI-assisted learning platforms, ensuring that they align with the principles of Universal Design for Learning (UDL). The Table 7 synthesizes the evaluation metrics, indicators, and methodological instruments identified across the reviewed studies. Organized according to the three UDL-aligned dimensions—accessibility, interactivity, and adaptability—it presents the empirical variables used in AI-based learning platforms to assess inclusiveness, usability, and user experience. Each entry integrates the nature of the variable (quantitative or mixed), the associated metrics (e.g., WCAG conformance, feedback accuracy, learning gains), the instruments employed in the studies, and representative authors. This structure provides a comprehensive map of how current research operationalizes the UDL principles through measurable technical and pedagogical criteria. Additionally, based on the metrics and indicators summarized in the previous table, the following rubric (Table 8 ) establishes three performance levels (High, Medium, and Low) for each UDL dimension. These levels reflect the degree to which AI-based learning platforms implement the empirical features documented in the literature, including technical compliance (e.g., WCAG, ISO/IEC), multimodal representation, bidirectional interaction, adaptive learning analytics, and validated personalization mechanisms. The rubric translates evidence-based indicators into observable criteria, allowing platforms to be classified according to how strongly they embody UDL principles in practice. Table 7 Summary of variables, metrics, and instruments for evaluating the AI–UDL Platform UDL Dimension Variable / Subdimension Nature Indicators / Metrics Instruments / Techniques Key References Accessibility (Representation) Regulatory and technical compliance Quantitative WCAG 2.1 (A/AA/AAA); ISO/IEC 25022; ISO 9241 − 210 (effectiveness, efficiency, satisfaction) WCAG/WCAG-EM audits; ISO checklists; task-based tests (Barbu et al., 2025 ; Kenneth et al., 2024 ; Maksymov et al., 2025 ) Multimodal and linguistic accessibility Quantitative / Mixed Captioning accuracy, sign recognition rate, multimodal coverage AI integration (ASR, TTS, NLP); validation with deaf users (Han et al., 2024 ; Memari & Taheri, 2024 ; Rodríguez-Moreno et al., 2023 ) Inclusive usability Quantitative SUS, UEQ, error rate, navigation clarity, learning time SUS/UEQ surveys; usability tests (Isaeva et al., 2025 ; Morgado et al., 2025 ) Ethics, privacy, and institutional support Qualitative / Mixed Data policies, digital equity, institutional support Document analysis, interviews, observations (Madanchian & Taherdoost, 2025 ; Mehmood, 2025 ; Morgado et al., 2025 ) Interactivity (Action and Expression) Cognitive dialogue and adaptive feedback Quantitative Feedback accuracy, response time, RMSE, explanatory quality Interaction analytics; feedback evaluation (Khosravi et al., 2023 ; Tan et al., 2025 ; Tebourbi et al., 2025 ) AI-mediated social interactivity Quantitative / Mixed Engagement, participation level, network and sentiment analysis Participation analytics; collaborative observation (Barbu et al., 2025 ; Khosravi et al., 2023 ; Sajja et al., 2025 ) Interactive usability and system responsiveness Quantitative Response time, error rate, CI, SQ Performance tests; TAM/TAM3 surveys (Han et al., 2024 ; Maksymov et al., 2025 ) Adaptability (Engagement and Motivation) Dynamic content and difficulty adjustment Quantitative Learning gains, RMSE, adaptation time, adaptive sequencing Fuzzy logic, pre/post-tests, performance traces (Kenneth et al., 2024 ; Memari & Taheri, 2024 ; Tan et al., 2025 ) Learner modeling Quantitative Accuracy/F1, BKT, profile updates, cognitive-affective variables MAS/RAG; log mining; learner sourcing (Demartini et al., 2024 ; Khosravi et al., 2023 ) Sustained motivation and engagement Mixed PU, PE, CI, CSE (TAM3), retention, affective signals TAM3, motivational analytics, emotion detection (Han et al., 2024 ; Mehmood, 2025 ; Sajja et al., 2025 ) Equity and diversity in adaptation Qualitative / Quantitative Fairness, bias detection, inclusion, multilingual support Algorithmic audits, interviews, stratified tests (Madanchian & Taherdoost, 2025 ; Sheejamol et al., 2025 ) Table 8 Rubric for Evaluating AI-Based Learning Platforms According to UDL Dimensions UDL Dimension High Level Medium Level Low Level Accessibility (Representation) • Implements multiple means of representation (sign language, text, audio, animations, graphics). • Fully complies with WCAG 2.1 (A/AA/AAA), ISO 25022, and ISO 9241. • Provides multimodal accessibility validated with deaf users (ASR, TTS, NLP, sign recognition). • High usability metrics: SUS/UEQ in positive ranges, low error rates, clear navigation. • Addresses ethics, privacy, digital equity, and institutional support. • Accessibility features are present but only partially or inconsistently implemented. • Partial WCAG compliance (e.g., only Level A) or standards mentioned without evidence. • Moderate usability (average SUS scores, recurring errors). • Multimodal accessibility limited (e.g., only captions or only text). • Accessibility is addressed only theoretically, without technical implementation. • Does not comply with WCAG or ISO standards. • Offers a single mode of representation (e.g., text only). • No usability metrics or validation with deaf learners. Interactivity (Action and Expression) • Provides real bidirectional interaction: sign-language avatars, immediate feedback, multimodal responses. • High feedback precision: high accuracy, low response time, low RMSE. • Applies participation analytics, network analysis, and sentiment analysis. • Intelligent tutoring or feedback systems aligned with learner performance. • Interaction exists but is limited to basic inputs (clicks, buttons) with minimal personalization. • Feedback is delayed or generic. • Only basic social interaction available (simple chat or comments). • Limited reporting of feedback metrics (CI, SQ, error rates). • No functional interactivity. • Bidirectionality is mentioned but not implemented. • No dynamic feedback mechanisms; interaction is static. • No interactivity metrics (e.g., TAM, TAM3, CI). Adaptability (Engagement and Motivation) • Fully adaptive platforms: real-time adjustment of content, routes, difficulty, and sequencing. • Uses predictive analytics, BKT, cognitive–affective modeling. • Shows consistent learning gains, low RMSE, and adaptive performance traces. • Adjusts pace and vocabulary dynamically (continual learning). • High retention and sustained personalization with affective-state detection. • Adaptation is mentioned but not fully implemented. • Basic adjustments (pace or difficulty) but reactive rather than predictive. • No cognitive–affective profiling. • Learning gains inconsistent or not reported. • Personalization limited or rule-based. • No personalization. • Static platform with fixed routes and non-adaptive content. • No adaptive analytics, learner modeling, or dynamic sequencing. • No evaluation using RMSE, BKT, learning gains, or personalization metrics. 6 Discussion The systematic analysis of recent literature reveals that evaluating AI-assisted learning platforms cannot be considered solely from each UDL dimension independently. Rather, their value emerges from the synergistic interaction between accessibility, interactivity, and adaptability, forming a comprehensive educational ecosystem. The convergence of these dimensions allows technology not only to facilitate access but also to transform the learning experience into a dynamic, inclusive, and sustainable process. Regarding the first research question (how the evaluation of AI platforms has been addressed) it is observed that efforts tend to focus on technical personalization and adaptability. However, a cross-dimensional perspective shows that the effectiveness of adaptability critically depends on accessibility and interactivity: AI models can offer tailored learning paths, but these will only be meaningful if students can perceive, interact with, and receive feedback in comprehensible and contextually appropriate formats (Han et al., 2024 ; Sajja et al., 2025 ). This interdependence suggests that measuring each dimension in isolation may underestimate the actual effects on learning and inclusion. Concerning the second question, which metrics and frameworks are used, the review indicates that normative, perceptual, and algorithmic instruments, although useful, work best when integrated holistically. Usability and user experience metrics (SUS, UEQ) should be linked with adaptive and affective performance indicators, while technical standards (WCAG, ISO) are complemented by real-time learning and interaction data. This integration allows for the assessment of not only technical effectiveness but also the pedagogical and ethical quality of the platform, providing concrete evidence of how AI can facilitate active participation and deep learning. Regarding the third question, how findings can guide inclusive design based on UDL, the results suggest that the synergy between dimensions offers a reference framework for ethical and pedagogical design decisions. Accessibility ensures entry points, interactivity transforms the experience into meaningful engagement, and adaptability guarantees personalization and sustainability. The integration of cross-dimensional metrics allows the identification of gaps in the student experience, adjustment of pedagogical interventions, and monitoring of algorithmic fairness. In this way, UDL principles go beyond mere regulatory compliance and materialize into effective and equitable learning pathways. Finally, the discussion highlights tensions and opportunities for future research. While AI can act as a catalyst for inclusion, there remains a need for teacher oversight, data governance, and algorithmic transparency (Kenneth et al., 2024 ; Tebourbi et al., 2025 ). True inclusion depends on the platforms’ ability to coherently integrate accessibility, interactivity, and adaptability, rather than merely on technological sophistication. This perspective also aligns with the Sustainable Development Goals (SDG 4 and 9), suggesting that educational AI can contribute to equity and innovation if conceived within a comprehensive techno-pedagogical framework. 7 Implications for Future Research The findings of this review open several lines of research aimed at strengthening the convergence between UDL and AI in education, from a perspective that integrates technical rigor, pedagogical sense, and ethical commitment. First, there is a clear need to develop comprehensive evaluation models that combine normative metrics, such as ISO/IEC 25022 and WCAG 2.1 guidelines, with pedagogical and affective indicators capable of simultaneously assessing technical accessibility, user experience, and learning outcomes. The literature still lacks standardized frameworks linking software quality to educational quality; advancing empirically validated mixed instruments constitutes a priority challenge for future research. A second line points to expanding the application contexts of UDL. Most reviewed studies are confined to higher education, leaving basic education scenarios and populations with functional or linguistic diversity underexplored. It will be essential to design and evaluate platforms that incorporate linguistic accessibility (such as automatic sign language translation or gesture recognition) and multimodal adaptations that respond to neurodiversity. Such research can provide evidence on how UDL principles translate into tangible inclusive practices, particularly in learning environments for deaf students or those with visual impairments. Third, it is necessary to compare the effectiveness of currently employed metrics. Studies rely on usability, engagement, or algorithmic accuracy indicators without establishing their correspondence or cross-validation. Future research should advance toward mixed methodologies that integrate statistical analyses, interaction data mining, and qualitative exploration of the learning experience, so that technical metrics can be interpreted in terms of their pedagogical relevance. A fourth emerging line involves transparency and algorithmic governance in intelligent educational environments. Few studies examine how to audit AI to detect bias, ensure data privacy, or make personalization processes understandable. Exploring participatory auditing mechanisms, open data policies, or algorithmic equity analyses by student subgroups represents a fertile area for interdisciplinary research. Furthermore, it will be valuable to investigate in greater depth the relationship between emotional interactivity and motivational adaptability. Integrating affective sensors, multimodal analytics, or neuro educational models would allow researchers to understand how emotions influence personalization effectiveness, paving the way for more empathetic, learner-centered systems. Finally, future research should reassess the role of teachers in AI educational ecosystems. Results suggest that automation without pedagogical mediation can lead to personalization devoid of educational meaning. Studies on human-AI collaboration is needed to examine how teachers can guide, supervise, and enrich system decisions, ensuring coherence between pedagogical objectives and algorithmic adaptations. More broadly, this review calls for an epistemological shift: moving from research focused on technical efficiency toward a deeper understanding of comprehensive, inclusive, and ethical evaluation of AI in education. Advancing evidence on how accessibility, interactivity, and adaptability metrics translate into genuine learning, well-being, and equity will be the next necessary step to consolidate a new field of study: universal evaluation of AI-assisted learning. Declarations Conflict of interest The authors declare that they have no known financial or non-financial competing interests that could have appeared to influence the work reported in this paper. Funding The authors declare that no funds, grants, or other financial support were received during the preparation of this manuscript. Author Contribution Frank Alexander Parra conceived the study, defined the research objectives, and led the design of the systematic review. He coordinated the development of the analytical framework based on Universal Design for Learning (UDL) and supervised the construction of the evaluation metrics and rubrics presented in the study. Juan Pablo Rivera contributed to the literature search strategy, screening process, data extraction, and qualitative synthesis of the reviewed studies. He also participated in the analysis of evaluation metrics and in the drafting and revision of the manuscript. Both authors independently screened the retrieved studies in multiple rounds (title and abstract, full text, and methodological appraisal) following the inclusion and exclusion criteria. Disagreements were resolved through discussion and consensus. Inter-coder reliability was calculated using Cohen’s κ (κ = 0.84), indicating substantial agreement. The academic advisors provided methodological guidance, critical feedback on the analytical framework, and reviewed the manuscript for conceptual coherence and academic rigor. All authors read and approved the final manuscript. Data Availability The datasets generated and analyzed during the current study, including the coding matrix and extracted data from the reviewed articles, are available from the corresponding author upon reasonable request. References Adomavicius, G., Tuzhilin, A.: Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-Art and Possible Extensions. IEEE Trans. Knowl. Data Eng. 17 (6), 734–749 (2005). https://doi.org/10.1109/TKDE.2005.99 Ara, J., SikLanyi, C., Kelemen, A.: Accessibility engineering in web evaluation process: a systematic literature review. Univ. Access Inf. Soc. 23 (2), 653–686 (2024). https://doi.org/10.1007/s10209-023-00967-2 Barbu, M., Iordache, D.D., Petre, I., Barbu, D.C., Băjenaru, L.: Framework Design for Reinforcing the Potential of XR Technologies in Transforming Inclusive Education [Article]. Appl. Sci. (Switzerland). 15 (3) (2025). Article 1484. https://doi.org/10.3390/app15031484 Barkovska, O., Ruban, I., Tymoshenko, D., Holovchenko, O., Yankovskyi, O., RESEARCH ON MOBILE MACHINE LEARNING PLATFORMS FOR HUMAN GESTURE RECOGNITION IN HUMAN-MACHINE INTERACTION SYSTEMS [Article]: Technol. Audit Prod. Reserves. 2 (82), 6–14 (2025). https://doi.org/10.15587/2706-5448.2025.325423 Barsalou, L.W.: Language comprehension: Archival memory or preparation for situated action. Discourse Processes. 28 (1), 61–80 (1999) Bercaru, V., Popescu, N.: A Systematic Review of Accessibility Techniques for Online Platforms: Current Trends and Challenges. Applied Sciences , 14 (22). (2024) Botella, P., Burgués, X., Carvallo, J., Franch, X., Grau, G., Marco, J., Quer, C.: ISO/IEC 9126 in practice: what do we need to know? (2004) CAST: Universal Design for Learning Guidelines version 2.2. In. (2018) Demartini, C.G., Sciascia, L., Bosso, A., Manuri, F.: Artificial Intelligence Bringing Improvements to Adaptive Learning in Education: A Case Study. Sustainability, 16 (3). (2024) El-Sabagh, H.A., Hamed, S.M.: Adaptive E-learning Systems Based on Learning Styles: A UDL Perspective. Int. J. Educational Technol. High. Educ. 18 , 1–22 (2021). https://doi.org/10.1186/s41239-021-00289-4 Gee, J.P.: Reading as Situated Language: A Sociocognitive Perspective. J. Adolesc. Adult Lit. 44 (8), 714–725 (2001) Halkiopoulos, C., Gkintoni, E.: Leveraging AI in E-Learning: Personalized Learning and Adaptive Assessment through Cognitive Neuropsychology—A Systematic Analysis. Electronics , 13 (18). (2024) Han, J., Liu, G., Liu, X.: Continue using or gathering dust? A mixed method research on the factors influencing the continuous use intention for an AI-powered adaptive learning system for rural middle school students. Heliyon. 10 (12), e33251 (2024). https://doi.org/10.1016/j.heliyon.2024.e33251 Herrera Nieves, L., Moya, C., E., Montes Soldado, R., ASSESSMENT METHODOLOGY USING THE UNIVERSAL DESIGN FOR LEARNING PERSPECTIVE: Turkish Online J. Distance Educ. 26 (3), 238–255 (2025). https://doi.org/10.17718/tojde.1510242 MOODLE USABILITY Herrera Nieves, L.B., Moya, C., E., Soldado, M., R: Moodle Usability Assessment Methodology using the Universal Design for Learning perspective. In: Anadolu University (2025) Hssina, B., Erritali, M.: A Personalized Pedagogical Objectives Based on a Genetic Algorithm in an Adaptive Learning System. Procedia Comput. Sci. 151 , 1152–1157 (2019). https://doi.org/https://doi.org/10.1016/j.procs.2019.04.164 Ifenthaler, D., Gibson, D.: Adoption of Data Analytics in Higher Education Learning and Teaching. Springer (2020). https://doi.org/10.1007/978-3-030-48884-0 Ingavélez-Guerra, P., Otón-Tortosa, S., Hilera-González, J., Sánchez-Gordón, M.: The use of accessibility metadata in e-learning environments: a systematic literature review. Univ. Access Inf. Soc. 22 (2), 445–461 (2023). https://doi.org/10.1007/s10209-021-00851-x Isaeva, R., Karasartova, N., Dznunusnalieva, K., Mirzoeva, K., Mokliuk, M., THROUGH ADAPTIVE LEARNING PLATFORMS AND EMERGING COMPUTER TECHNOLOGIES IN EDUCATION: Jurnal Ilmiah Ilmu Terapan Universitas Jambi. 9 (1), 144–160 (2025). https://doi.org/10.22437/jiituj.v9i1.37967 ENHANCING LEARNING EFFECTIVENESS Ismailova, R., Inal, Y.: Comparison of Online Accessibility Evaluation Tools: An Analysis of Tool Effectiveness. IEEE Access. 10 , 58233–58239 (2022). https://doi.org/10.1109/ACCESS.2022.3179375 ISO: ISO 9241 – 210:2019 Ergonomics of human-system interaction — Part 210: Human-centred design for interactive systems . (2019). https://www.iso.org/standard/77520.html ISO, & IEC: ISO/IEC 25022: Systems and Software Engineering — Systems and Software Quality Requirements and Evaluation (SQuaRE) — Measurement of Quality in Use. ISO/IEC, In. Geneva, Switzerland (2016) Kenneth, H., Marino, M., Vasquez, E., Taub, M., Hunt, J., Tazi, Y.: Navigating AI-Powered Personalized Learning in Special Education: A Guide for Preservice Teacher Faculty. J. Special Educ. Preparation. 4 (2), 90–95 (2024). https://doi.org/10.33043/5b2xqcb3 Khosravi, H., Denny, P., Moore, S., Stamper, J.: Learnersourcing in the age of AI: Student, educator and machine partnerships for content creation. Computers Education: Artif. Intell. 5 , 100151 (2023). https://doi.org/https://doi.org/10.1016/j.caeai.2023.100151 Landis, J.R., Koch, G.G.: The Measurement of Observer Agreement for Categorical Data. Biometrics. 33 (1), 159–174 (1977). https://doi.org/10.2307/2529310 Lu, J.: Artificial Intelligence in Education Design and Evaluation of Adaptive Learning Systems. Region - Educational Res. Reviews. 6 (4), 209–214 (2024) Madahana, M., Khoza-Shangase, K., Moroe, N., Mayombo, D., Nyandoro, O., Ekoru, J.: A proposed artificial intelligence-based real-time speech-to-text to sign language translator for South African official languages for the COVID-19 era and beyond: In pursuit of solutions for the hearing impaired [Article]. South Afr. J. Commun. Disord. 69 (2) (2022). Article a915 https://doi.org/10.4102/sajcd.v69i2.915 Madanchian, M., Taherdoost, H.: Examining Critical Factors in Selecting AI Tools for Educational Success. Procedia Comput. Sci. 263 , 923–933 (2025). https://doi.org/https://doi.org/10.1016/j.procs.2025.07.111 Maksymov, M., Davydov, V., Petik, T., Grishyn, M.: Development of an Assessment Methodology for Educational Platforms in the Context of Educators’ Innovation and Research Activities. Filosofiya osvity Philos. Educ. 31 , 90–110 (2025). https://doi.org/10.31874/2309-1606-2025-31-1-6 Mehmood, H.: Artificial intelligence in educational governance: Implications for adaptive learning and student mindset formation. Rev. Educ. Adm. Law. 8 (2), 343–359 (2025). https://doi.org/10.47067/real.v8i2.429 Memari, M., Taheri, A.: Adaptive Teaching of the Iranian Sign Language Based on Continual Learning Algorithms. IEEE Access. 12 , 164164–164177 (2024). https://doi.org/10.1109/ACCESS.2024.3492056 Montes, A., Jiménez, R., García, F.: Evaluating Moodle Usability from a UDL Perspective: A Case Study in Latin American Higher Education. Educ. Inform. Technol. 29 (3), 455–478 (2024) Morgado, E., Leonido, L., Pereira, A., Gouveia, L.B.: Technology-Mediated Education: impact of AI on the main distance learning modalities [Article]. Educational Process: Int. J. 16 (2025). Article e2025211 https://doi.org/10.22521/edupij.2025.16.211 Obeid, A., Ibrahim, R., Fadhil, A.: Extended Model of Expectation Confirmation Model to Examine Users’ Continuous Intention Toward the Utilization of E-Learning Platforms. IEEE Access. 12 , 40752–40764 (2024). https://doi.org/10.1109/ACCESS.2024.3373190 Page, M.J., McKenzie, J.E., Bossuyt, P.M., Boutron, I., Hoffmann, T.C., Mulrow, C.D., Moher, D.: The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 372 , n71 (2021). https://doi.org/10.1136/bmj.n71 Parody, L.M., Leiva, J.-J., Santos-Villalba, M.J.: El Diseño Universal para el Aprendizaje en la Formación Digital del Profesorado desde una Mirada Pedagógica Inclusiva. Revista Latinoam. de educación inclusiva. 16 , 109–123 (2022). https://doi.org/10.4067/S0718-73782022000200109 Parveen, K., Obidallah, W.J., Alghamdi, A.A., Alduraywish, Y.A., Shafiq, M.: Gesture-enhanced adaptive learning platform for personalized AI-driven education. Interact. Learn. Environ., 1–22. https://doi.org/10.1080/10494820.2025.2542897 Punitham, M., Balasubramaniam, P., Lakshmi, S., Sriram, N., Shivakumar, K.M.: Navigating open and distance learning trends, advantages, challenges and innovations for the future education - A systematic review. Plant. Sci. Today. 12 (2) (2025). https://doi.org/10.14719/pst.8108 Rajabi, K.I.: Beyond the LMS: How a Multi-Platform, AI-Enhanced Ecosystem Boosts Motivation, Interaction, Deep Learning, and Design Competence among Arabic-Speaking Graduate Students. (2025). https://doi.org/10.21203/rs.3.rs-7156104/v1 Rodríguez-Moreno, I., Martínez-Otzeta, J.M., Sierra, B.: HAKA: HierArchical Knowledge Acquisition in a sign language tutor. Expert Syst. Appl. 215 , 119365 (2023). https://doi.org/https://doi.org/10.1016/j.eswa.2022.119365 Saborío-Taylor, A., Rojas-Ramírez, G.: Universal design for learning and artificial intelligence in the digital era: Fostering inclusion and autonomous learning. Int. J. Prof. Dev. Learners Learn. 6 (2), ep2408 (2024). https://doi.org/10.30935/ijpdll/14694 Saborío-Taylor, M., Rojas-Ramírez, J.: Universal design for learning and artificial intelligence in the digital era: Fostering inclusion and autonomous learning. Int. J. Prof. Dev. Learners Learn. 6 (2), ep2408 (2024). https://doi.org/10.30935/ijpdll/14694 Sajja, R., Sermet, Y., Cwiertny, D., Demir, I.: Integrating AI and Learning Analytics for Data-Driven Pedagogical Decisions and Personalized Interventions in Education. Technology, Knowledge and Learning . (2025). https://doi.org/10.1007/s10758-025-09897-9 Sheejamol, P.T., Chacko, A.M., Kumar, S.D.M.: Beyond the One-Size-Fits-All: A Systematic Review of Personalized and Gamified e-Learning for Neurodivergent Learners [Article]. Electron. J. e-Learning. 23 (3), 101–119 (2025). https://doi.org/10.34190/ejel.23.3.4051 Slepankova, M., Kilianova, K., Kockova, P., Kostolanyova, K., Kotyrba, M., Habiballa, H.: Student Perceptions and Preferences in Personalized AI-driven Learning [Article]. Acta Informatica Pragensia. 14 (2), 261–271 (2025). https://doi.org/10.18267/j.aip.278 Tan, L.Y., Hu, S., Yeo, D.J., Cheong, K.H.: Artificial intelligence-enabled adaptive learning platforms: A review. Computers Education: Artif. Intell. 9 , 100429 (2025). https://doi.org/https://doi.org/10.1016/j.caeai.2025.100429 Tebourbi, H., Nouzri, S., Mualla, Y., Abbas-Turki, A.: Artificial Intelligence Agents for Personalized Adaptive Learning. Procedia Comput. Sci. 265 , 252–259 (2025). https://doi.org/https://doi.org/10.1016/j.procs.2025.07.179 UNESCO: Consenso de Beijing sobre la inteligencia artificial y la educación . (2019). https://unesdoc.unesco.org/ark:/48223/pf0000368303 VanLehn, K.: The Behavior of Tutoring Systems. Int. J. Artif. Intell. Educ. 16 (3), 227–265 (2006) Wangdi, T., Shimray, R.: AI-Powered ReadTheory as a Self-Access Learning Platform to Enhance EFL Learners’ Reading Enjoyment and Comprehension Skills: A Posthumanist Perspective [Article]. SiSal J. 16 (2), 437–460 (2025). https://doi.org/10.37237/160209 Zapata, C., Acosta, J.: Innovaciones Tecnológicas Para Inclusión Educativa De Estudiantes Sordos. Ingeniería e Innovación. 6 (2) (2019). https://doi.org/10.21897/23460466.1595 Zhang, Y., Han, Y., Zhu, Z., Jiang, X.: Artificial intelligence in sign language recognition: A comprehensive bibliometric and visual analysis [Article]. Comput. Electr. Eng. 120 , 109854 (2024). https://doi.org/10.1016/j.compeleceng.2024.109854 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8801285","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Systematic Review","associatedPublications":[],"authors":[{"id":587264082,"identity":"6e870038-b05f-47ff-8e4d-1f6ca17fbdf3","order_by":0,"name":"Frank Alexander Parra Sánchez","email":"data:image/png;base64,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","orcid":"","institution":"Politécnico Colombiano Jaime Isaza Cadavid","correspondingAuthor":true,"prefix":"","firstName":"Frank","middleName":"Alexander Parra","lastName":"Sánchez","suffix":""},{"id":587264083,"identity":"8a8bf690-23e4-4471-8bd8-9dd37cbd2aae","order_by":1,"name":"Juan Pablo Rivera Barrera","email":"","orcid":"","institution":"Politécnico Colombiano Jaime Isaza Cadavid","correspondingAuthor":false,"prefix":"","firstName":"Juan","middleName":"Pablo Rivera","lastName":"Barrera","suffix":""},{"id":587264084,"identity":"96511380-a02b-4a6a-bc34-fd15994d6d98","order_by":2,"name":"Henry Omar Sarmiento Maldonado","email":"","orcid":"","institution":"Politécnico Colombiano Jaime Isaza Cadavid","correspondingAuthor":false,"prefix":"","firstName":"Henry","middleName":"Omar Sarmiento","lastName":"Maldonado","suffix":""},{"id":587264085,"identity":"a4dfb5f6-0c7f-4507-bf7f-db2c05ffca73","order_by":3,"name":"Carmen Amalia Camacho Sanabria","email":"","orcid":"","institution":"Caro and Cuervo Institute","correspondingAuthor":false,"prefix":"","firstName":"Carmen","middleName":"Amalia Camacho","lastName":"Sanabria","suffix":""}],"badges":[],"createdAt":"2026-02-05 23:53:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8801285/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8801285/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104399008,"identity":"bdc48cd2-2c53-47be-b38c-e885dbd3cb29","added_by":"auto","created_at":"2026-03-11 12:04:29","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":84548,"visible":true,"origin":"","legend":"\u003cp\u003eFlow diagram of the article selection process.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8801285/v1/385c6d9d69b28d253fe2dd4b.jpg"},{"id":103553112,"identity":"6ced57bc-cd46-4666-b549-e8d5e7675364","added_by":"auto","created_at":"2026-02-27 03:02:13","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":36897,"visible":true,"origin":"","legend":"\u003cp\u003ePercentage distribution of studies by dimension.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8801285/v1/07f3d9f6762d0b37ce101cbb.jpg"},{"id":104398161,"identity":"1e40cc7d-4396-421a-95b0-6d0ab611b29e","added_by":"auto","created_at":"2026-03-11 12:00:03","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":37709,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of studies according to target audience.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8801285/v1/178da68a910d0bb4b8a3e860.jpg"},{"id":104398969,"identity":"5c9f7f7a-4129-42f5-8988-57fba77e9f01","added_by":"auto","created_at":"2026-03-11 12:04:25","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":38929,"visible":true,"origin":"","legend":"\u003cp\u003eCo-occurrence of accessibility, interactivity, and adaptability across the analyzed studies.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8801285/v1/6edc29b1b1890f3068af877a.jpg"},{"id":109296546,"identity":"e038d9d1-7c53-4abc-89c5-338af7c6ab76","added_by":"auto","created_at":"2026-05-15 08:48:03","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":630522,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8801285/v1/e3e0144d-bb4a-464d-adbe-cc991c7b6966.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"AI-Based Learning Platforms: A Systematic Review of Evaluation Metrics for Accessibility, Interactivity and Adaptability through the Lens of Universal Design for Learning","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eArtificial Intelligence (AI) has increasingly shaped contemporary educational technologies, redefining the ways in which digital platforms are designed, evaluated, and personalized. Recent advances in learning analytics, adaptive algorithms, and intelligent feedback systems have enabled the development of environments capable of responding dynamically to learners\u0026rsquo; profiles, behaviors, and performance patterns (Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Within this landscape, the UDL framework has consolidated its role as one of the most influential approaches for promoting inclusion in technology-mediated education. By proposing multiple means of representation, action and expression, and engagement, UDL provides a pedagogical foundation to design equitable environments that accommodate diverse needs from the outset (CAST, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAlthough AI and UDL have evolved largely in parallel, their intersection presents an emerging opportunity: integrating data-driven adaptivity with principles of accessibility and inclusive design. This convergence is especially relevant in a context where digital learning platforms increasingly mediate instruction, assessment, and interaction. However, despite rapid progress in AI-enhanced learning systems, the literature still lacks systematic analyses that connect these technologies with UDL-based evaluative criteria. Existing reviews have largely concentrated on algorithmic performance, adaptive learning architectures, or technological innovation (such as learning analytics, recommender systems, or intelligent tutoring systems) without examining how these systems align with pedagogical principles of accessibility, interactivity, and adaptability (Khosravi et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Tan et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Tebourbi et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThis disjointed focus has produced a methodological gap: most AI-driven educational platforms are evaluated through computational indicators (accuracy, RMSE, precision/recall), while human-centered dimensions such as usability, accessibility compliance, satisfaction, and learner engagement remain secondary (ISO \u0026amp; IEC, 2016; Montes et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In other words, although AI systems increasingly influence learning processes, the frameworks used to assess them rarely consider whether they support inclusive pedagogical practices or embody the flexibility promoted by UDL.\u003c/p\u003e \u003cp\u003eTo address this limitation, the present study conducts a systematic review of evaluation metrics and assessment frameworks applied to AI-based learning platforms, following the PRISMA guidelines and examining peer-reviewed studies published from 2019 onward. UDL is used as the conceptual and analytical scaffold, enabling the organization of findings around three operational dimensions derived from the framework: accessibility (representation), interactivity (action and expression), and adaptability (engagement and motivation). This conceptual structure allows the review to analyze not only how platforms are technically evaluated, but also how current metrics reflect (or fail to reflect) the pedagogical commitments of inclusive design (Barbu et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Ingav\u0026eacute;lez-Guerra et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Mehmood, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAccordingly, this review seeks to map the state of the art in evaluating AI-driven learning platforms and to propose an integrative model that brings together international standards (e.g., ISO, WCAG), user-experience indicators, and adaptive performance measures. By bridging technical methodologies and pedagogical frameworks for inclusion, the study contributes to building evaluative criteria capable of guiding the development of more equitable, transparent, and accessible AI-based educational systems.\u003c/p\u003e \u003cp\u003eSpecifically, this review addresses the following research questions:\u003c/p\u003e \u003cp\u003eHow has recent literature evaluated AI-based learning platforms in terms of accessibility, interactivity, and adaptability?\u003c/p\u003e \u003cp\u003eWhat metrics, instruments, or reference frameworks are used to measure these dimensions?\u003c/p\u003e \u003cp\u003eHow can these findings inform the design of inclusive learning environments based on UDL principles?\u003c/p\u003e \u003cp\u003eThe article is organized as follows: Section 2 outlines the theoretical framework connecting UDL principles with international quality metrics for educational software. Section 3 describes the PRISMA-based methodology used for study identification, selection, and analysis. Section 4 presents the results structured around the three operational dimensions of UDL. Finally, Section 5 discusses the implications of these findings for evaluating and designing inclusive AI-driven learning platforms.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"2 Related works","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eRecent research in educational technology reveals a fragmented yet rapidly evolving convergence between AI, accessibility standards, and inclusive pedagogy. Existing systematic reviews and empirical studies cluster around four dominant lines.\u003c/p\u003e \u003cp\u003eThe first concerns digital accessibility and quality assurance, emphasizing persistent inconsistencies between international frameworks (such as WCAG 2.1 and ISO/IEC 25022) and their actual implementation in educational platforms. Automated auditing tools often produce divergent outcomes for the same resources, particularly in STEM environments, where accessibility barriers in navigation, contrast, and non-text content remain unresolved (Ismailova \u0026amp; Inal, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Zapata \u0026amp; Acosta, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). These limitations are compounded by educators\u0026rsquo; limited awareness of accessibility and UDL principles.\u003c/p\u003e \u003cp\u003eA second body of literature focuses on AI-driven personalized learning and adaptive assessment, where platforms increasingly employ machine learning algorithms (such as Fuzzy C-Means, Random Forest, or Deep Knowledge Tracing) to infer cognitive states and adapt content dynamically (Lu, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Adaptive Learning Platforms (ALPs) and intelligent tutoring systems demonstrate significant gains in knowledge retention, yet often prioritize prediction accuracy over inclusivity and transparency (Parveen et al.).\u003c/p\u003e \u003cp\u003eThe third line addresses multimodal and gesture-based interfaces for inclusive interaction. Speech, text, and gesture recognition systems have enhanced accessibility for diverse learners, though continuous sign-language recognition remains a major technical challenge requiring robust datasets and cross-linguistic validation (Barkovska et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Bercaru \u0026amp; Popescu, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Finally, a growing strand examines accessibility metadata in open and distance learning (ODL). Despite the existence of structured standards such as IMS AfA V3.0 and ISO/IEC 24751, their integration into MOOCs and e-learning systems remains limited (Ingav\u0026eacute;lez-Guerra et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), while learner satisfaction and persistence continue to depend heavily on perceived accessibility and usability (Obeid et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Punitham et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTaken together, these studies illustrate remarkable technological progress but also expose a persistent methodological gap: the absence of an integrative framework that connects accessibility, interactivity, and adaptability, the operational core of UDL. Although prior reviews have examined aspects of AI-enhanced education, none has systematically analyzed the evaluation metrics linking these three dimensions in the context of AI-based learning platforms. Addressing this gap, the present study proposes a unified approach that bridges normative quality standards (WCAG, ISO/IEC 25022), algorithmic performance indicators, and user-experience analytics to advance a coherent model for evaluating inclusive, AI-driven learning environments.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"3 Theoretical framework","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThis section presents the conceptual foundations that guide the analysis of AI-based learning platforms from an inclusive and evaluative perspective. The framework draws on three complementary components: (a) UDL as the pedagogical foundation for inclusion; (b) AI as the technological infrastructure enabling personalization and automation; and (c) international quality standards (ISO/IEC and WCAG) as reference models for assessing software usability, accessibility, and effectiveness. Together, these perspectives provide an integrated lens to understand how intelligent learning environments can be evaluated not only for their technical performance but also for their capacity to promote equitable, accessible, and adaptive learning.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Fundations of UDL\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe UDL framework is one of the most influential approaches to inclusive education in contemporary pedagogy. Developed by (CAST, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), UDL is based on the premise that learner diversity is inherent to any educational process, and therefore instructional design must anticipate variability rather than respond to it retroactively. UDL proposes three core principles:\u003c/p\u003e \u003cp\u003e \u003cem\u003eMultiple means of representation\u003c/em\u003e (the \u003cem\u003ewhat\u003c/em\u003e of learning), which promote offering information in varied perceptual and linguistic formats;\u003c/p\u003e \u003cp\u003e \u003cem\u003eMultiple means of action and expression\u003c/em\u003e (the \u003cem\u003ehow\u003c/em\u003e of learning), which broaden the ways students interact with content and demonstrate understanding; and\u003c/p\u003e \u003cp\u003e \u003cem\u003eMultiple means of engagement\u003c/em\u003e (the \u003cem\u003ewhy\u003c/em\u003e of learning), which sustain motivation, interest, and persistence.\u003c/p\u003e \u003cp\u003eThese principles align with broader global commitments to equitable and quality education, including Sustainable Development Goal 4 (UNESCO, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). In digital environments, UDL reframes diversity as a design criterion rather than as an instructional constraint. This perspective has been increasingly used to guide the development of digital materials, pedagogical strategies, and formative assessment practices (Parody et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; A. Sabor\u0026iacute;o-Taylor \u0026amp; G. Rojas-Ram\u0026iacute;rez, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). When digital learning platforms incorporate adaptive technologies, UDL becomes especially pertinent: its principles can be operationalized through multimodal interfaces, personalized feedback, and algorithmic adjustments that support learner variability.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e3.2 AI and UDL: Toward Inclusive Intelligent Learning Environments\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eAI-based learning platforms have introduced new possibilities for personalization, including recommendation systems, adaptive difficulty adjustment, conversational tutors, and predictive analytics. These systems can analyze learner interactions and adjust instructional pathways accordingly. However, technological sophistication alone does not guarantee inclusiveness; systems may optimize performance metrics without supporting equitable participation or diverse learning needs (Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe integration of AI and UDL thus represents an opportunity to redesign digital learning with both personalization and inclusion in mind. While UDL provides the pedagogical foundation for removing barriers through flexible representation, expression, and engagement, AI provides the technical mechanisms to enact these principles at scale and in real time (M. Sabor\u0026iacute;o-Taylor \u0026amp; J. Rojas-Ram\u0026iacute;rez, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This convergence transforms UDL from a design philosophy into an evaluative lens: the behavior of the system (its feedback, adaptivity, and multimodal mediation) can be assessed in terms of how well it reflects UDL\u0026rsquo;s commitments to accessibility, interactivity, and learner-centered engagement (El-Sabagh \u0026amp; Hamed, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.3 From UDL Principles to the Operational Dimensions of Accessibility, Interactivity, and Adaptability\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eAlthough UDL is formulated as a pedagogical framework, its principles can be translated into operational dimensions that enable empirical evaluation in digital environments. This translation is consistent with the sociocognitive view of learning presented by Gee, who argues that comprehension is grounded in perceptual, embodied, and contextual processes rather than in the decoding of linguistic symbols alone. According to Gee, understanding emerges through \u0026ldquo;perceptual simulations that prepare agents for situated action\u0026rdquo; (Barsalou, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e1999\u003c/span\u003e; Gee, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2001\u003c/span\u003e) and through \u0026ldquo;experiences (perceptions, feelings, actions, and interactions) stored as dynamic images tied to perception\u0026rdquo; (Gee, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). These ideas align directly with UDL\u0026rsquo;s emphasis on offering multiple means of representation, reinforcing accessibility as the dimension that ensures diverse perceptual and linguistic pathways for understanding information.\u003c/p\u003e \u003cp\u003eThe principle of multiple means of action and expression corresponds to the dimension of interactivity, which captures the extent to which learners can act upon and with the system. Gee highlights that learning is inseparable from speaking, listening, and interacting, with the material and social world, underscoring that meaning is constructed through embodied interaction rather than passive reception. Moreover, he notes that language scaffolds the performance of action in the world (Gee, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2001\u003c/span\u003e), supporting the relevance of feedback-rich, dialogic, and manipulable interfaces-core characteristics of interactivity in digital learning platforms.\u003c/p\u003e \u003cp\u003eFinally, the principle of engagement in UDL maps onto adaptability, understood as the system\u0026rsquo;s capacity to support motivation, persistence, and autonomy through personalized learning conditions. Gee states that meaning is always customized to our actual contexts, purposes, values, and intended courses of action, emphasizing that learners interpret information according to their goals and prior experiences. He explains this adaptive process through the metaphor of running our tapes (Gee, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2001\u003c/span\u003e), where learners draw on previous embodied experiences to engage with new situations. This notion parallels the adaptive mechanisms in AI-based platforms that adjust content, support, and difficulty levels in response to learner profiles and evolving engagement.\u003c/p\u003e \u003cp\u003eTaken together, Gee\u0026rsquo;s sociocognitive perspective supports the operationalization of UDL principles into the dimensions of accessibility, interactivity, and adaptability by showing that learning is inherently perceptual, interactive, and context-dependent. These dimensions therefore provide a coherent and empirically grounded structure for evaluating whether AI-based learning environments reflect the inclusive intentions of UDL in their design and behavior.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.4 International Standards (ISO/IEC and WCAG) as Complementary Evaluation Frameworks\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eBuilding on the UDL-derived dimensions described above, intelligent learning environments must also adhere to established technical quality standards that guide usability, accessibility, reliability, and \u0026ldquo;quality in use.\u0026rdquo; The ISO/IEC 9126 and ISO/IEC 25000 SQuaRE series offer widely recognized frameworks for software evaluation, including effectiveness, efficiency, satisfaction, and freedom from risk (Botella et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; ISO \u0026amp; IEC, 2016). These standards have been applied extensively in the evaluation of educational software and digital interfaces, providing measurable criteria for verifying user performance in real-world contexts (ISO, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSimilarly, the Web Content Accessibility Guidelines (WCAG 2.1) define criteria for perceptibility, operability, understandability, and robustness in online environments, establishing minimum accessibility levels for inclusive digital content (W3C 2018). Research on accessibility technologies for deaf learners has demonstrated the value of aligning digital resources with WCAG to ensure equitable interaction, especially when sign language or multimodal interfaces are involved (Saunders et al. 2019; Huenerfauth et al. 2008).\u003c/p\u003e \u003cp\u003eWhen combined with UDL, ISO and WCAG frameworks offer complementary evaluative structures:\u003c/p\u003e \u003cp\u003e \u003cem\u003eAccessibility\u003c/em\u003e aligns with WCAG and ISO effectiveness/efficiency criteria.\u003c/p\u003e \u003cp\u003e \u003cem\u003eInteractivity\u003c/em\u003e corresponds to ISO measurements of productivity, error recovery, and system responsiveness.\u003c/p\u003e \u003cp\u003e \u003cem\u003eAdaptability\u003c/em\u003e links to ISO constructs such as reliability, flexibility, and satisfaction, which capture system responsiveness to user variability (Ifenthaler \u0026amp; Gibson, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; ISO, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThus, ISO and WCAG standards reinforce UDL\u0026rsquo;s pedagogical commitments with verifiable technical metrics.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.5 An Integrative Framework for the Evaluation of Intelligent Learning Environments\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eBringing these components together, UDL principles, AI capabilities, and international evaluation standards, creates a coherent analytical model for understanding inclusive intelligent learning environments. Prior work has emphasized the need for frameworks that bridge pedagogical intent with technological implementation in AI-driven systems (El-Sabagh \u0026amp; Hamed, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; VanLehn, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). In this study, UDL defines what inclusive systems must guarantee (accessibility, meaningful interaction, and adaptive engagement), while AI provides the mechanisms to enact these commitments through multimodal mediation, personalization algorithms, and intelligent feedback (Adomavicius \u0026amp; Tuzhilin, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; M. Sabor\u0026iacute;o-Taylor \u0026amp; J. Rojas-Ram\u0026iacute;rez, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). ISO/IEC and WCAG standards, in turn, offer measurable and internationally validated criteria for assessing usability, accessibility, and quality in use.\u003c/p\u003e \u003cp\u003eThis integrated framework underpins the analytical model applied in the systematic review. It guides the identification, classification, and interpretation of evaluation metrics reported in AI-based learning platforms and supports the broader goal of advancing inclusive, transparent, and pedagogically aligned digital learning environments (L. Herrera Nieves et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Ingav\u0026eacute;lez-Guerra et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Mehmood, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTo summarize these connections, Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e links the three UDL-derived dimensions with their corresponding ISO/IEC and WCAG quality metrics. This synthesis provides a concise framework for evaluating inclusiveness, usability, and adaptability in AI-based learning environments.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCorrespondence between UDL principles, operational dimensions, and ISO/WCAG quality metrics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDimension\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMain purpose\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIn UDL\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eIn ISO/WCAG (software quality)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePredominant variable type\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccessibility\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnsuring that all users can access the learning environment and understand the information.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRepresentation principle.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eISO 9241\u0026thinsp;\u0026minus;\u0026thinsp;210 \u0026rarr; perceptibility, usability; WCAG 2.1 \u0026rarr; perceivable, operable, understandable, robust; ISO 25022 \u0026rarr; effectiveness, efficiency, context coverage.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eStructural (entry conditions).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteractivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnabling the user to act on the system and receive meaningful, timely feedback.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAction and expression principle.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eISO 9126-4 / ISO 25010 \u0026rarr; productivity, response time, feedback effectiveness, error recovery.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eProcessual (use dynamics).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaptability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdjusting the experience to the user (profile, progress, motivation, or cognitive/affective state).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEngagement principle.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eISO 25022 / ISO 9126 \u0026rarr; satisfaction, reliability, flexibility, freedom from risk.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePersonalizing (intelligent adjustment).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4 Methodology","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThis study followed the Preferred Reporting Items for Systematic Reviews and Meta- Analyses (PRISMA 2020) guidelines (Page et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) to ensure methodological transparency and reproducibility. The review process comprised four sequential stages: (1) definition of the research question and eligibility criteria, (2) systematic search in indexed databases, (3) screening and coding of retrieved studies, and (4) qualitative and quantitative synthesis of evidence.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Study design\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe review adopted an exploratory\u0026ndash;descriptive design aimed at identifying and analyzing evaluation metrics applied to AI-based learning platforms through the principles of UDL. The protocol was registered internally and validated by two domain experts before execution, guaranteeing alignment between the research questions and the coding categories derived from the theoretical framework (accessibility, interactivity, and adaptability).\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Search Strategy and Information Sources\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe search was conducted between 2024 and 2025 across six major databases: Scopus, IEEE Xplore, Web of Science, ScienceDirect, SpringerLink, and ACM Digital Library. Boolean search equations were tailored to each database to maximize precision and recall. Representative strings included:\u003c/p\u003e \u003cp\u003e(\"AI-based learning platform*\" OR \"intelligent tutoring system*\" OR \"adaptive learning system*\") AND (\"evaluation metric*\" OR \"assessment framework*\" OR \"usability\" OR \"learning analytic*\") AND (\"accessibility\" OR \"interactivity\" OR \"adaptability\") AND\u003c/p\u003e \u003cp\u003e(\"Universal Design for Learning\" OR \"UDL\") Only peer-reviewed articles published between 2019 and 2025 were considered, reflecting the post-pandemic acceleration of AI adoption in education. Additional records were retrieved through backward snowballing from the reference lists of key studies.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Selection and data-coding process\u003c/h2\u003e \u003cp\u003eAll retrieved documents were exported to Zotero for deduplication and then screened independently by two reviewers in three rounds: title/abstract, full text, and methodological appraisal. Eligibility followed the inclusion/exclusion criteria summarized in Table 2. Disagreements were resolved through discussion and consensus. To guarantee reliability, intercoder agreement was calculated using Cohen\u0026rsquo;s \u0026kappa;, yielding \u0026kappa; = 0.84, which indicates substantial agreement according to \u0026nbsp;(Landis \u0026amp; Koch, 1977). This statistical control reinforces the consistency of decisions across reviewers and minimizes subjectivity in classification.\u0026nbsp;\u003c/p\u003e\u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003cp\u003eTable 2 Correspondence between UDL principles, operational dimensions, and ISO/WCAG quality metrics\u003c/p\u003e\u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInclusion Criteria (IC)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExclusion Criteria (EC)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIC1.\u003c/b\u003e Studies describing or evaluating educational platforms assisted by artificial intelligence.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eEC1.\u003c/b\u003e Studies describing platforms without the use of AI or related techniques.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIC2.\u003c/b\u003e Research addressing at least one of the three components: accessibility, interactivity, or adaptability.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eEC2.\u003c/b\u003e Studies focused solely on pedagogical content without considering technical aspects of the platform.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIC3.\u003c/b\u003e Publications between 2018 and 2025, peer-reviewed, in English or Spanish.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eEC3.\u003c/b\u003e Publications prior to 2018 or non-scientific articles (e.g., posters, blogs, etc.).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIC4.\u003c/b\u003e Articles describing technical functionalities or AI tools implemented in educational environments.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eEC4.\u003c/b\u003e Purely theoretical studies without application or validation of the proposed technology.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIC5.\u003c/b\u003e Open-access articles.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eEC5.\u003c/b\u003e Articles with access limitations.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Data extraction and coding scheme\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eA structured coding matrix was developed based on the analytical dimensions of UDL and on the ISO/IEC 25022 quality-in-use indicators. Each article was coded across 20 variables, including publication year, country, educational level, AI techniques, evaluation metrics, accessibility compliance, and type of data used. The coding process was supported by Atlas.ti 23, which facilitated the identification of co-occurring themes and the visualization of conceptual clusters. Memos and code families were employed to track the relationships between AI methods and UDL principles.\u003c/p\u003e \u003cp\u003eThe final corpus comprised 24 studies meeting all inclusion criteria. Each record was assigned a unique identifier and cross-checked against the database source to prevent duplication.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.5 Analysis and synthesis of information\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe synthesis combined descriptive statistics (frequency and co-occurrence distributions) with qualitative thematic analysis of conceptual and methodological trends. Quantitative data were processed in Python 3.12 using pandas and matplotlib libraries to compute frequency tables and figures. Qualitative interpretation focused on how evaluation frameworks operationalized the UDL dimensions of accessibility, interactivity, and adaptability.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the PRISMA flow diagram summarizing identification, screening, eligibility, and inclusion stages.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.6 Use of generative artificial intelligence tools\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eDuring the preparation of this manuscript, the authors made limited use of generative artificial intelligence tools exclusively to support language editing and stylistic refinement. These tools were employed solely to improve clarity, coherence, and grammatical accuracy of the text, without contributing to the study design, data analysis, interpretation of results, or generation of substantive scientific content. All content was critically reviewed and edited by the authors, who assume full responsibility for the accuracy, originality, and integrity of the published work.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"5 Results","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e5.1 General Characteristics\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe analysis of the 24 selected articles made it possible to identify, from the perspective of UDL principles, how recent literature addresses the evaluation of AI-based learning platforms, focusing on the criteria of accessibility, interactivity, and adaptability.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the proportion of studies that address each of these dimensions. Adaptability is the most frequently explored (42%), followed by interactivity (33%) and accessibility (25%). This pattern confirms that research attention is concentrated on the personalization and algorithmic adjustment of the learning experience, while universal accessibility components remain less explored.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eRegarding the educational level of application (see Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e), most of the studies are situated in higher education contexts (13 articles), followed by mixed-level studies (4 articles) and, to a lesser extent, experiences in primary and secondary education (2 articles). Seven studies do not specify the educational level of implementation. This trend highlights a significant gap in research targeting early school levels, where UDL principles could have a particularly relevant impact on inclusion and motivation for learning.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents the combinations of features analyzed. Most studies do not focus on a single dimension but rather integrate multiple dimensions simultaneously. The most frequent combination is adaptability\u0026thinsp;+\u0026thinsp;interactivity (11 articles), followed by the integration of all three dimensions (8 articles). In contrast, only two studies focus exclusively on a single category (one on accessibility and one on adaptability). This finding reflects a clear trend toward multidimensional approaches, in which personalization and interaction converge, while accessibility continues to play a complementary rather than a structural role.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eRegarding the evaluation metrics, the analyzed studies can be grouped into three categories: international standards and guidelines (WCAG 2.1, ISO 25022, ISO 9241); usability and technology acceptance models (SUS, TAM, UTAUT, UEQ); and algorithmic and performance metrics (efficiency, effectiveness, error rate, engagement, feedback accuracy, precision/recall, RMSE). Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents this classification, establishing the relationship between the evaluation metrics, their purposes, their correspondence with UDL dimensions, and the authors who support them.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClassification of metrics and standards used in the reviewed studies\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReference group\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExamples of metrics or models\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEvaluation focus / UDL mapping (with sources)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInternational standards and guidelines\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWCAG 2.1; ISO/IEC 25022:2016; ISO 9241\u0026ndash;210\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEstablish technical criteria for accessibility, usability, and software quality (effectiveness, efficiency, user satisfaction, flexibility). This aligns mainly with accessibility (representation) and adaptability (efficiency, flexibility). Reported in (Ara et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; L. Herrera Nieves et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Ingav\u0026eacute;lez-Guerra et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; ISO \u0026amp; IEC, 2016).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUsability and technology acceptance models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSUS (System Usability Scale); TAM (Technology Acceptance Model); UTAUT (Unified Theory of Acceptance and Use of Technology); UEQ (User Experience Questionnaire)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEvaluate user perception, ease of use, satisfaction, acceptance, and learning experience in AI-mediated environments. These instruments map to interactivity (action and expression) and adaptability (engagement, autonomy, satisfaction). (Zhang et al., \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlgorithmic and performance metrics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEfficiency, effectiveness, error rate, \u003cem\u003eengagement\u003c/em\u003e, \u003cem\u003efeedback accuracy\u003c/em\u003e, \u003cem\u003eprecision/recall\u003c/em\u003e, RMSE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasure AI algorithm performance, personalization effectiveness, and the system\u0026rsquo;s predictive capability in real time. These metrics correspond to adaptability (dynamic adjustment and adaptive learning) and interactivity (personalized feedback and participation analysis). Reported in (Demartini et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Isaeva et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Mehmood, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe convergence between international standards (such as ISO/IEC 25022 and WCAG 2.1), technology acceptance models (TAM, UTAUT, SUS), and algorithmic performance metrics (efficiency, precision, error rate, and engagement) reveals a clear trend toward the triangulation of technical, cognitive, and pedagogical approaches in the evaluation of AI-based educational platforms. In the reviewed literature, these metrics do not operate in isolation; rather, they are articulated in a complementary manner to assess quality, usability, and learning impact in alignment with the UDL principles.\u003c/p\u003e \u003cp\u003eThis landscape suggests that evaluation processes are shifting from a merely functional perspective of software toward an integrated assessment of the educational experience, where technical performance, student engagement, and personalized learning are analyzed as interdependent dimensions. Nevertheless, the results also highlight gaps in the systematization of the applied criteria: while ISO standards and WCAG guidelines provide standardized indicators of accessibility and quality in use, machine learning metrics and user perception scales tend to operate in a fragmented way, without an explicit alignment with UDL principles.\u003c/p\u003e \u003cp\u003eThe following section deepens the analysis by categories, examining how the reviewed studies address the three critical dimensions that operationalize UDL principles (accessibility, interactivity, and adaptability) identifying the most representative metrics, techniques, and approaches that support their empirical evaluation.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Analysis by Category\u003c/h2\u003e \u003cdiv id=\"Sec19\" class=\"Section3\"\u003e \u003ch2\u003e5.2.1 Accessibility\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eAccessibility represents the dimension most closely aligned with the UDL principle of multiple means of representation, ensuring that all users (regardless of their sensory, cognitive, or contextual conditions) can effectively perceive, understand, and navigate digital environments. In AI-assisted platforms, this principle is operationalized through inclusive design strategies, international quality standards, and assistive technologies aimed at eliminating or reducing barriers to information access.\u003c/p\u003e \u003cp\u003eAmong the 24 studies analyzed, only 25% address accessibility as a central focus, confirming its lower prevalence compared to adaptability and interactivity. However, the studies that do incorporate it approach accessibility through three complementary perspectives: (1) compliance with international technical standards, (2) integration of assistive and multimodal accessibility tools, and (3) perceptual evaluation of inclusive usability.\u003c/p\u003e \u003cp\u003eThe first group of studies associates accessibility with the structural foundation of digital inclusion, including verification of compliance with international standards (WCAG 2.1, ISO 9241, ISO/IEC 25022) and the incorporation of governance, ethics, and privacy principles as part of inclusive design. (Maksymov et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), for example, address accessibility through a quantitative assessment of technical compliance. Their proposed methodology includes explicit metrics on the availability of accessibility features and support programs (technical, financial, or technological), translating regulatory compliance into measurable equity indicators. In a more normative dimension, (Madanchian \u0026amp; Taherdoost, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Mehmood, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) broaden the concept toward ethics and governance, emphasizing that digital equity also depends on data privacy, user security, and the objectivity of AI-based systems. These perspectives combine technical compliance metrics with trust and transparency indicators, aligning accessibility with its role as a digital right.\u003c/p\u003e \u003cp\u003eThe second group of studies conceptualizes accessibility as a dynamic condition in which AI acts as a mediator to provide multiple and sensorially adaptive representations. (Memari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) exemplify this approach with their adaptive sign language teaching model for Iranian Sign Language, using metrics such as word weight, repetition, speed, and user score to adjust content difficulty based on user performance, applying fuzzy logic to personalize the experience. Similarly, (Sheejamol et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) work with neurodivergent populations (ASD, ADHD, dyslexia) and demonstrate that adaptive, multisensory gamification improves engagement and knowledge retention, reinforcing the link between accessibility and cognitive adaptability. Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e expand the concept into Extended Reality (ER) environments, integrating voice, text, and gesture modalities to create immersive and accessible educational experiences. Although their metrics focus on task completion rates, their overall objective is to transform accessibility into an immersive and participatory experience, consistent with UDL\u0026rsquo;s representation principle. Likewise, (Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) emphasize the role of teachers in multimodal inclusion, underscoring that accessibility requires not only adaptive technologies but also pedagogical competencies to mediate their implementation. Their study highlights the need for teacher training in inclusive AI tools (e.g., OneNote, Socratic, Quizlet), positioning accessibility as a formative (not merely technical) process.\u003c/p\u003e \u003cp\u003eThe third approach combines user experience assessment with the institutional support needed to ensure equitable access and the sustainability of inclusive design. (Maksymov et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), for instance, consider navigation speed and number of clicks as efficiency indicators, incorporating institutional support programs as an additional criterion and extending the evaluation to the organizational dimension. Other studies, such as those by (Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) and (Rajabi, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), frame usability within Scientific and UX Design, measuring affective states such as stress, curiosity, and confusion to adapt personalized interventions, while integrating indicators of robust technical infrastructure and system acceptance as key elements of user satisfaction. Finally, (Morgado et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) and (Isaeva et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) highlight the importance of teacher and institutional support, emphasizing continuous teacher training and material diversity as essential conditions to sustain accessibility and meaningful learning.\u003c/p\u003e \u003cp\u003eBased on the above, the findings indicate that accessibility in AI-based platforms should not be understood merely as the availability of technical resources, but rather as an integrated construct that encompasses regulatory compliance, multimodal inclusion, and inclusive usability. In this way, accessibility ceases to be an isolated starting point and becomes a cross-cutting indicator of inclusive quality, which supports the subsequent dimensions of the model: interactivity and adaptability.\u003c/p\u003e \u003cp\u003eThis perspective is complemented by the identification of a set of principles or requirements that the authors, either explicitly or implicitly, highlight in their studies and that should be considered when designing accessible learning platforms, along with the guiding questions to achieve this. Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e summarizes this information\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePrinciples, requirements, and guiding questions for accessible platforms\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor and year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrinciples or requirements (explicit or implicit)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGuiding questions for accessible design\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Maksymov et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccessibility and inclusivity as a universal evaluation criterion; adherence to WCAG and international standards; evaluation of mobile accessibility; institutional support for inclusion.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDoes the platform meet accessibility requirements (WCAG) for users with visual, auditory, motor, or cognitive disabilities? Does the platform provide text descriptions for videos and images?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Barbu et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCompliance with accessibility standards in ER; WCAG 2.2 and W3C ER Accessibility User Requirements; generation of AI-driven content adaptation for students with special educational needs.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDoes the design framework adhere to WCAG 2.2 and W3C ER Accessibility User Requirements to ensure inclusion? Does the system dynamically adapt learning environments to each learner\u0026rsquo;s needs and progress?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUDL to dynamically adjust content, interaction, and engagement; legal compliance (Section 508 and WCAG); need for teacher training to balance technologies and promote inclusion.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAre the accessibility criteria stipulated by WCAG and Section 508 being met through AI-driven personalization? Are educators being trained to integrate AI-based technologies and balance automation with pedagogical oversight?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCognitive accessibility for neurodivergent learners (ASD, ADHD, dyslexia); mitigation of sensory overload; multimodal strategies (text, visual, audio, haptic); integration with UDL.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIs the UI/UX design minimizing cognitive overload in neurodivergent learners? Does the system adapt to individual cognitive profiles instead of forcing a one-size-fits-all model?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Memari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eContinuous and multimodal adaptation for sign languages (e.g., Iranian Sign Language); continual learning to expand vocabulary without catastrophic forgetting; fuzzy logic for personalized teaching parameters (playback speed, repetition, scoring).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDoes the system adapt modality parameters (speed, repetition) to each learner\u0026rsquo;s performance capacity? How does the AI system manage vocabulary recognition accuracy while continuously expanding its lexicon?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Rajabi, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRobust technical infrastructure and high-quality UX as success factors for adaptation across diverse populations; multimodal prompts.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIs the technological framework robust and user-friendly enough to support personalized and adaptive learning for diverse student populations?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Mehmood, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEthical governance and inclusivity; ensuring that AI systems are fair and understandable.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAre the AI systems fair, understandable, and inclusive? Do governance principles prevent malpractices and protect the rights of all students?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Morgado et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDigital inclusion and equity of access regardless of socioeconomic or geographic context.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHave concrete strategies been implemented to guarantee digital inclusion and platform accessibility for all students?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section3\"\u003e \u003ch2\u003e5.2.2 Interactivity (UDL Principle of Action and Expression)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eInteractivity constitutes one of the most critical operational pillars for understanding how AI\u0026ndash;based platforms foster active student participation in learning. In alignment with the second principle of UDL, providing multiple means of action and expression, this category examines the capacity of intelligent environments to enable bidirectional communication, stimulate critical thinking, and provide meaningful feedback.\u003c/p\u003e \u003cp\u003eThe reviewed studies show that interactivity, when mediated by AI algorithms, is structured around three key subdimensions: (1) cognitive interactivity, which fosters reasoning and self-regulation; (2) social interactivity, which enables collaboration and peer learning; and (3) algorithmic interactivity, which enhances the user experience through automated feedback, response pattern analysis, and task personalization.\u003c/p\u003e \u003cp\u003eThe first subdimension focuses on the accuracy and personalization of feedback that the system provides to the learner, functioning as an intelligent tutor that adjusts guidance and difficulty in real time according to the user\u0026rsquo;s performance, cognitive states, and affective states. In this line, (Memari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) propose an adaptive teaching model for Iranian Sign Language (ISL), applying fuzzy logic and continual learning to adjust parameters such as word weight, repetition, and sign speed. This architecture enables effective and profile-sensitive interaction, achieving a significant improvement in perceived adaptability (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.63).\u003c/p\u003e \u003cp\u003eSimilarly, (Tan et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) integrate the Revised Bloom\u0026rsquo;s Taxonomy (RBT) into intelligent tutoring systems, automatically adjusting the difficulty level of tasks and questions to sustain students\u0026rsquo; cognitive progression, while (Khosravi et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) develop the RiPPLE system, which uses AI to provide personalized hints and explanations in peer learning activities (learnersourcing). This approach turns feedback into a formative process of self-regulation and collective knowledge construction.\u003c/p\u003e \u003cp\u003e(Tebourbi et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) introduce the concept of dynamic scaffolding through the AIA-PAL framework, based on Multi-Agent Systems (MAS) and Large Language Models (LLMs). This model adjusts support levels as learners progress, ensuring a balance between challenge and guidance. Complementarily, (Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) examine AI-Powered Personalized Learning (AI-PPL) and its capacity to keep learners within their zone of proximal development through real-time feedback in tools such as Socratic.\u003c/p\u003e \u003cp\u003eOther studies, such as (Wangdi \u0026amp; Shimray, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) and (Slepankova et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), emphasize the importance of immediate feedback in sustaining engagement and com- prehension. However, they caution that superficial or shallow feedback can limit cognitive depth. (Halkiopoulos \u0026amp; Gkintoni, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) advocate for intelligent feedback mechanisms based on cognitive neuropsychology principles, which adjust content to the user\u0026rsquo;s thinking style and emotional state.\u003c/p\u003e \u003cp\u003eThe second subdimension explores the ability of AI to facilitate collaboration and peer communication, promoting social learning and community building in virtual environments. In this regard, (Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) examines emotional interaction and its relationship with engagement through VirtualTA, a tool that uses GPT-4 and natural language processing to identify affective states such as stress, curiosity, confusion, and agitation. This emotional profiling enables personalized interventions that sustain collective motivation.\u003c/p\u003e \u003cp\u003eIn the domain of collaborative learning, (Khosravi et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) report that the RiPPLE system increases trust in peer assessment by reducing bias through identity anonymization and reinforcing the perception of fairness. Only 2% of users expressed disagreement with the scores received, demonstrating the model\u0026rsquo;s reliability. Likewise, (Barbu et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) extend social interaction to immersive ER environments, where virtual agents or NPCs adapt their responses and tone to the user\u0026rsquo;s communicative style, fostering social skill development in inclusive contexts. (Isaeva et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) highlight the role of simulations, virtual labs, and educational games in active learning, while (Sheejamol et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) show that collaborative gamification strengthens communication and negotiation among neurodivergent students, promoting a participatory and inclusive environment.\u003c/p\u003e \u003cp\u003eThe final subdimension focuses on technical fluency and UX-factors that determine the perceived quality and sustainability of interaction. (Maksymov et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) identify usability and interface design as central criteria in the evaluation of educational platforms, prioritizing simplicity, minimalism, and reducing the number of steps required to complete a task. (Han et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) link System Quality (SQ) to continued use intention, showing that the fluidity of interactive functions and logical consistency are essential to maintain user motivation.\u003c/p\u003e \u003cp\u003eFrom a technical perspective, (Tebourbi et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) measure agent response time and workflow synchronization, demonstrating that delays or poorly timed transitions reduce the perceived naturalness of interaction. (L. B. Herrera Nieves et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), in their evaluation of Moodle, emphasize the importance of clear organization, a comprehensible interface, and functional aesthetics, confirming that perceived usability is an essential component of interactivity.\u003c/p\u003e \u003cp\u003e(Madahana et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) argue that usability should be intuitive and accessible to non-expert users, enabling teachers to integrate AI tools without requiring advanced technical knowledge. Finally, (Rajabi, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) and (Demartini et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) highlight the relationship between robust technical infrastructure and learning effectiveness, demonstrating that environments with high interactivity and low cognitive load foster user satisfaction and retention.\u003c/p\u003e \u003cp\u003eOverall, the studies indicate that interactivity in AI-driven platforms has evolved from simple user\u0026ndash;system exchange toward cognitive, social, and emotionally adaptive communication, aligned with UDL\u0026rsquo;s principle of action and expression. The three identified requirements ((1) cognitive dialogue and adaptive feedback, (2) AI-mediated social interactivity, and (3) interactive usability and system responsiveness) constitute an articulated framework that integrates intelligent feedback, meaningful collaboration, and technical fluency.\u003c/p\u003e \u003cp\u003eInteractivity, when mediated by AI, not only optimizes learning efficiency but also redefines learner agency as active participation in a human\u0026ndash;algorithmic interaction network. In this context, feedback quality, computational empathy, and user experience become critical indicators of equity and participation, positioning interactivity as an essential vector of inclusion and engagement in contemporary educational environments.\u003c/p\u003e \u003cp\u003eIn the analyzed studies, interactivity emerges as a balancing component between automation and human agency. The most successful platforms do not replace student participation but rather amplify it through intelligent mediations that integrate dialogue, adaptability, and collaboration. The identified metrics make it possible to quantify the quality of human\u0026ndash;machine interaction, while triangulation with ISO standards (particularly 25022 and 9241) provides a technical structure to measure user efficiency and satisfaction. Based on these ideas, these approaches demonstrate that interactivity constitutes the connection point between the pedagogical principles of UDL and the algorithmic capabilities of AI.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e synthesizes the principles or requirements identified around interactivity and the design questions formulated by the authors to guide the development of AI-based learning environments from pedagogical, communicative, and technological perspectives.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePrinciples, requirements, and guiding questions for the design of interactive platforms\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor and Year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrinciples or Requirements (explicit or implicit)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGuiding Questions for Interactive Design\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Memari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEffective Human\u0026ndash;Agent Interaction (HAI); adaptive teaching programs based on user performance; fuzzy logic to adjust system parameters (speed, repetition).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDoes the adaptive teaching system interact effectively with diverse users and achieve the desired adaptability in training sessions? How can learning parameters (playback speed, repetition) be optimized based on user performance (speed, accuracy, score)?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Wangdi \u0026amp; Shimray, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNeed for detailed explanatory feedback in Self-Access Language Learning (SALL) environments beyond multiple-choice interactions.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can SALL platforms be improved to provide detailed feedback and explanations for incorrect answers, reducing reliance on MCQs? How can limitations in learners\u0026rsquo; self-access experience be addressed through adaptive interactivity?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Demartini et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eValuable feedback and continuous content adaptation aligned with classroom dynamics; Intelligent Decision Support Systems (IDSS) for corrective actions (e.g., flipped classrooms, tutoring).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhat methods and tools are used to collect and process student data to inform interaction? What corrective actions can an IDSS provide to teachers or administrators?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Halkiopoulos \u0026amp; Gkintoni, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eContinuous, adaptive feedback mechanisms; use of engagement metrics to adjust learning experience in real time.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can attention and perception principles be used to create AI systems that personalize and enhance learning interaction? How can cognitive neuroscience help guide AI models toward more effective learning environments?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Han et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSQ, including recognition accuracy (e.g., in drawing questions) and logical difficulty progression; emphasis on perceived ease of use (PEU) and feedback.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can the adaptive engine be optimized to improve question logic and recognition accuracy, increasing users\u0026rsquo; Continuous Intention (CI) to use the system? Which perceived external control factors (teacher/technical support) are essential to sustain CI?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eAs a synthesis of this section, interactivity in AI-powered platforms emerges as a key dimension for materializing the UDL principle of action and expression. The reviewed studies agree that the quality of interaction depends not only on the number of exchanges but also on their cognitive and affective depth. The integration of intelligent agents, multimodal feedback, and emotional analysis enables the development of more autonomous, dialogic, and personalized learning experiences. However, ethical and pedagogical challenges remain regarding how to balance automation with human support.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e5.3 Adaptability (UDL Principle: Engagement and Involvement)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eAdaptability emerges as the most prominent category in the reviewed literature and functions as the central interface between AI and Universal UDL. In alignment with UDL\u0026rsquo;s third principle (providing multiple means of engagement and sustaining learner motivation) this dimension refers to the capacity of AI-driven systems to dynamically adjust content, strategies, and learning pathways in response to each learner\u0026rsquo;s characteristics, interests, and performance.\u003c/p\u003e \u003cp\u003eAcross studies, adaptability is not conceptualized as a static software property but rather as a continuous process of cognitive, emotional, and contextual adjustment requiring both algorithmic precision and intentional pedagogical design. Collectively, the literature identifies four core requirements that operationalize this dimension: dynamic adjustment of content and difficulty levels, learner modeling and predictive personalization, sustained motivation and engagement, and equity and diversity in adaptive processes. Regarding the first requirement, the evidence shows that adaptability functions through real-time modifications of content, learning pathways, and task complexity, supported by intelligent tutoring strategies and dynamic scaffolding. For example, (Tan et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) establish the conceptual basis of adaptive sequencing and assessment by measuring learning gains between pre- and post-tests. Their model applies the Revised Bloom\u0026rsquo;s Taxonomy (RBT) to adjust task difficulty and cognitive progression, enabling activities to evolve from lower-order skills to complex creative tasks.\u003c/p\u003e \u003cp\u003eIn an applied context, (Memari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) developed an adaptive platform for teaching Iranian Sign Language (ISL), using fuzzy logic and Elastic Weight Consolidation (EWC) to personalize the user experience. Their metrics (word weight, repetition frequency, and execution speed) allowed precise measurement of adaptive fit and interaction fluency, yielding a significant perceived adaptability effect (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.63). Likewise, (Hssina \u0026amp; Erritali, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) conceptualize adaptation as an optimization problem solvable through genetic algorithms that search for optimal learning pathways based on the distance between the learner\u0026rsquo;s current profile and target competencies. (Tebourbi et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) expand this approach through a MAS implementing dynamic scaffolding, progressively adjusting support levels in response to real-time interactions. Finally, (Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) emphasize that AI-PPL tools should deliver immediate adaptive feedback to sustain progression within each learner\u0026rsquo;s zone of proximal development.\u003c/p\u003e \u003cp\u003eThe second requirement positions learner modeling as the technical core of adaptability, integrating cognitive, behavioral, and emotional data to predict performance and inform pedagogical strategies. For instance, (Demartini et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) propose a closed-loop regulatory environment that dynamically updates learner profiles based on interaction data, enriched with indicators of socio-emotional competencies. This allows pedagogical decisions to be informed by a more holistic understanding of learning trajectories.\u003c/p\u003e \u003cp\u003e(Halkiopoulos \u0026amp; Gkintoni, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) argue that personalization should be guided by cognitive neuropsychology, incorporating cognitive styles and profiles into Adaptive Assessment (AA) systems. Empirical evidence from (Khosravi et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) supports this approach: their RiPPLE system leverages learner sourcing data to feed predictive models, demonstrating superior performance compared to systems relying solely on traditional assessments. These results highlight increased accuracy and relevance of personalized recommendations.\u003c/p\u003e \u003cp\u003eThe third requirement emphasizes the emotional dimension of adaptability, understood as the system\u0026rsquo;s ability to sustain learner motivation, self-efficacy, and positive attitudes through dynamic interventions. Drawing on the TAM3 model, (Han et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) identify Perceived Usefulness and Perceived Enjoyment as the most influential factors in CI to use adaptive environments, with Computer Self-Efficacy mediating the relationship between ease of use and persistence.\u003c/p\u003e \u003cp\u003eAffectively, (Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) developed a real-time emotion analysis system using GPT-4 to detect stress, curiosity, confusion, and agitation in learners\u0026rsquo; language. This continuous monitoring allows adaptive adjustment of both content and feedback types to sustain cognitive engagement. (Mehmood, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) extends this discussion to mindset formation, arguing that adaptive systems can strengthen not only academic performance but also positive learning attitudes and emotional resilience. These findings are reinforced by (Memari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), who report significant increases in learner motivation and willingness to interact with adaptive systems.\u003c/p\u003e \u003cp\u003eFinally, the fourth requirement suggests that adaptability must go beyond technical personalization and performance to ensure equity, inclusion, and algorithmic ethics, avoiding biases that perpetuate inequalities. (Sheejamol et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) argue that traditional one-size-fits-all e-learning models are ineffective for neurodivergent learners. They propose multimodal personalization and adaptive gamification models responsive to diverse cognitive styles and sensory needs. (Maksymov et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) reinforce this perspective by including adaptability to individual learning pace, multilingual localization, and institutional flexibility as measurable quality indicators. At an organizational level, (Madanchian \u0026amp; Taherdoost, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) emphasize that adaptability must be scalable and sustainable, ensuring effective AI implementation across different class sizes and educational contexts. (Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) and (L. B. Herrera Nieves et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) add that inclusive adaptation is achieved when both pedagogical and technical design incorporate UDL principles, ensuring accessibility for all learners.\u003c/p\u003e \u003cp\u003eAs in the previous sections, Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e presents the principles or requirements and the guiding questions based on the analyzed authors\u0026rsquo; works, which should be considered when designing user-adaptive platforms.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePrinciples, requirements, and guiding questions for designing adaptive platforms\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor \u0026amp; Year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrinciples or Requirements (explicit or implicit)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGuiding questions raised by the authors for adaptive design\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eHssina \u0026amp; Erritali, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAlgorithmic adaptability through genetic algorithms; personalized learning paths according to profile and objectives.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can learning paths evolve according to each student\u0026rsquo;s progress and learning style?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eHan et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAffective and cognitive adaptation through multimodal emotion recognition.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can difficulty and content be adjusted based on the learner\u0026rsquo;s emotional state and attention?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eRajabi, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdaptability based on machine learning and ISO standards; mobile and cognitive personalization.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can personalization enhance the experience without compromising fairness and transparency?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eTebourbi et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdaptability through cooperative agents (AIA-PAL); dynamic learning paths with human validation.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhat balance should exist between AI-driven automation and teacher supervision?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eMemari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLinguistic adaptability through continual learning (EWC); personalization of learning pace and vocabulary.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can cumulative learning be maintained without forgetting prior knowledge?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eSajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCognitive and emotional adaptability; personalized paths based on learner progress and affective states.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can AI trace itineraries that reflect students\u0026rsquo; cognitive and emotional development?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eWangdi \u0026amp; Shimray, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHierarchical adaptability (CEFR levels); automatic progression according to performance.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhat mechanisms allow the platform to evaluate and adjust the student\u0026rsquo;s level in real time?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eSheejamol et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGamified adaptability (PAGE model); adjustment of difficulty level and rewards according to learning style.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow can personalization, motivation, and performance be balanced for neurodivergent students?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003eBarbu et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSensory and cognitive adaptability in ER environments; 3D environment personalization according to performance and preference.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow should immersive environments respond to diverse learning paces and styles?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eBased on the analysis of the reviewed articles, adaptability emerges as the most developed dimension within AI-assisted educational platforms. The authors address this principle through algorithmic personalization, cognitive and affective adaptation, and continual learning, proposing systems that dynamically adjust content, difficulty, and pace according to student performance. These strategies aim to operationalize the UDL engagement principle by enabling flexible learning paths that are sensitive to individual differences. In summary, the literature shows that adaptability not only enhances personalization but also raises critical challenges related to equity, teacher supervision, and the sustainability of learning. While advances in machine learning and student modeling allow for immediate responses to individual needs, the challenge remains to ensure that these adaptations do not result in fragmented or decontextualized experiences, but rather strengthen autonomy, motivation, and the continuity of knowledge.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e5.4 Cross-sectional Analysis of the Three Dimensions\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eBased on the 24 articles included in the review, recent literature increasingly addresses the evaluation of AI-assisted learning platforms from a systemic perspective that integrates three interdependent dimensions of the UDL: accessibility (representation), interactivity (action and expression), and adaptability (engagement). Rather than being analyzed separately, these dimensions function as linked elements of a pedagogical quality cycle. Coverage analysis indicates that adaptability receives the most attention (42%), followed by interactivity (33%) and, to a lesser extent, accessibility (25%). Most studies focus on higher education contexts (12 studies), with fewer in primary/secondary education (2 studies), four mixed-scope proposals, and seven unspecified, highlighting a research gap in school stages where UDL could have a more direct impact.\u003c/p\u003e \u003cp\u003eThe convergence across dimensions is evident at multiple levels. Accessibility has evolved beyond mere regulatory compliance (WCAG 2.1, ISO 9241/ISO 25022) to integrate with interactivity and adaptability through AI-mediated supports, such as automatic captioning, sign language interpreters, multimodal interfaces, and ER accessibility features. These interventions ensure that accessible representation translates into meaningful action and active participation, promoting equity in learning.\u003c/p\u003e \u003cp\u003eInteractivity bridges access and personalization, turning available content into effective cognitive and social experiences. Adaptive feedback, dynamic tutoring, and AI-enabled social mediation (learner sourcing, scaffolding, and decision support ecosystems) enhance learning relevance and engagement. Interactive usability and technical fluency ensure that access opportunities are effectively utilized, while affective reading allows real-time adjustment according to students\u0026rsquo; emotional states.\u003c/p\u003e \u003cp\u003eAdaptability serves as the integrative core, linking technical personalization with pedagogical inclusion. Student modeling, adaptive sequencing, continuous learning, and real-time scaffolding optimize content difficulty and pacing while coordinating accessibility and interactivity. These adaptations address neurodiversity, multilingual needs, scalability, and the AI\u0026ndash;teacher balance, ensuring personalized experiences that are equitable and pedagogically meaningful.\u003c/p\u003e \u003cp\u003eEvaluation metrics converge across the three dimensions into a comprehensive framework:\u003c/p\u003e \u003cp\u003eStandards and regulations: WCAG 2.1, ISO 9241, ISO/IEC 25022, ensuring effectiveness, efficiency, context coverage, and usability.\u003c/p\u003e \u003cp\u003eUsability and acceptance models: SUS, UEQ, TAM/TAM3, UTAUT, capturing\u003c/p\u003e \u003cp\u003euser perceptions of ease-of-use, satisfaction, and engagement.\u003c/p\u003e \u003cp\u003eAlgorithmic and educational performance metrics: feedback accuracy, response time, precision/recall, RMSE, semantic analysis of interactions, pre/post Learning Gains, Bloom/RBT mapping, error and dropout rates, participation/collaboration measures, and real-time affective indicators.\u003c/p\u003e \u003cp\u003eIn ER environments and neurodivergent contexts, additional requirements address cognitive load, multisensory responses, and data governance, demonstrating the convergence of technical, pedagogical, and ethical criteria.\u003c/p\u003e \u003cp\u003eIn summary, the evidence indicates that AI-based learning platform design should operate systemically, leveraging the interplay of accessibility, interactivity, and adaptability through integrated metrics. This approach guides inclusive, adaptive, and student-centered designs, ensuring learning is meaningful, equitable, and motivating.\u003c/p\u003e \u003cp\u003eDesign Guidelines from UDL\u003c/p\u003e \u003cp\u003eEmpirical evidence suggests a concrete roadmap:\u003c/p\u003e \u003cp\u003eAccessibility as an entry condition: combine normative auditing (WCAG; ISO 9241/25022) with functional testing in diverse populations (e.g., deaf users), verifying effectiveness/efficiency, context coverage, and understandability (SUS/UEQ). Include linguistic accessibility (interpreters/sign avatars, captions) and cognitive accessibility (minimalist UI, load reduction).\u003c/p\u003e \u003cp\u003eInteractivity as pedagogical mediation: ensure explanatory and timely feedback (beyond correct/incorrect) with quality traceability; affective reading for personalized interventions; meaningful collaboration (peers/NPCs); and technical fluency (response times, coordination, low errors) to support student agency.\u003c/p\u003e \u003cp\u003eAdaptability as ethical and sustainable adjustment: operate with robust learner models (cognitive\u0026ndash;behavioral\u0026ndash;affective), adaptive sequencing/evaluation (Bloom/RBT), dynamic scaffolding (MAS/LLMs), and bias controls (multilingual, devices, subgroups) with teacher supervision and governance policies. Based on the analysis of the reviewed literature, it is possible to establish a roadmap to guide the design of AI-assisted learning platforms, ensuring that they align with the principles of Universal Design for Learning (UDL).\u003c/p\u003e \u003cp\u003eThe Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e synthesizes the evaluation metrics, indicators, and methodological instruments identified across the reviewed studies. Organized according to the three UDL-aligned dimensions\u0026mdash;accessibility, interactivity, and adaptability\u0026mdash;it presents the empirical variables used in AI-based learning platforms to assess inclusiveness, usability, and user experience. Each entry integrates the nature of the variable (quantitative or mixed), the associated metrics (e.g., WCAG conformance, feedback accuracy, learning gains), the instruments employed in the studies, and representative authors. This structure provides a comprehensive map of how current research operationalizes the UDL principles through measurable technical and pedagogical criteria.\u003c/p\u003e \u003cp\u003eAdditionally, based on the metrics and indicators summarized in the previous table, the following rubric (Table\u0026nbsp;\u003cspan refid=\"Tab8\" class=\"InternalRef\"\u003e8\u003c/span\u003e) establishes three performance levels (High, Medium, and Low) for each UDL dimension. These levels reflect the degree to which AI-based learning platforms implement the empirical features documented in the literature, including technical compliance (e.g., WCAG, ISO/IEC), multimodal representation, bidirectional interaction, adaptive learning analytics, and validated personalization mechanisms. The rubric translates evidence-based indicators into observable criteria, allowing platforms to be classified according to how strongly they embody UDL principles in practice.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of variables, metrics, and instruments for evaluating the AI\u0026ndash;UDL Platform\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUDL Dimension\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVariable / Subdimension\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eIndicators / Metrics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eInstruments / Techniques\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eKey References\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccessibility (Representation)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRegulatory and technical compliance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWCAG 2.1 (A/AA/AAA); ISO/IEC 25022; ISO 9241\u0026thinsp;\u0026minus;\u0026thinsp;210 (effectiveness, efficiency, satisfaction)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWCAG/WCAG-EM audits; ISO checklists; task-based tests\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Barbu et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Maksymov et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMultimodal and linguistic accessibility\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative / Mixed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCaptioning accuracy, sign recognition rate, multimodal coverage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAI integration (ASR, TTS, NLP); validation with deaf users\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Han et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Memari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Rodr\u0026iacute;guez-Moreno et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInclusive usability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSUS, UEQ, error rate, navigation clarity, learning time\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSUS/UEQ surveys; usability tests\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Isaeva et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Morgado et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEthics, privacy, and institutional support\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQualitative / Mixed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eData policies, digital equity, institutional support\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDocument analysis, interviews, observations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Madanchian \u0026amp; Taherdoost, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Mehmood, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Morgado et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteractivity (Action and Expression)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCognitive dialogue and adaptive feedback\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFeedback accuracy, response time, RMSE, explanatory quality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eInteraction analytics; feedback evaluation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Khosravi et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Tan et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Tebourbi et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAI-mediated social interactivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative / Mixed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEngagement, participation level, network and sentiment analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eParticipation analytics; collaborative observation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Barbu et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Khosravi et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInteractive usability and system responsiveness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eResponse time, error rate, CI, SQ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePerformance tests; TAM/TAM3 surveys\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Han et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Maksymov et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaptability (Engagement and Motivation)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDynamic content and difficulty adjustment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLearning gains, RMSE, adaptation time, adaptive sequencing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFuzzy logic, pre/post-tests, performance traces\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Memari \u0026amp; Taheri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Tan et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLearner modeling\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAccuracy/F1, BKT, profile updates, cognitive-affective variables\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMAS/RAG; log mining; learner sourcing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Demartini et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Khosravi et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSustained motivation and engagement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMixed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePU, PE, CI, CSE (TAM3), retention, affective signals\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTAM3, motivational analytics, emotion detection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Han et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Mehmood, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEquity and diversity in adaptation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQualitative / Quantitative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFairness, bias detection, inclusion, multilingual support\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAlgorithmic audits, interviews, stratified tests\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(Madanchian \u0026amp; Taherdoost, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Sheejamol et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab8\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRubric for Evaluating AI-Based Learning Platforms According to UDL Dimensions\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUDL Dimension\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh Level\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedium Level\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLow Level\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccessibility (Representation)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull; Implements multiple means of representation (sign language, text, audio, animations, graphics). \u0026bull; Fully complies with WCAG 2.1 (A/AA/AAA), ISO 25022, and ISO 9241. \u0026bull; Provides multimodal accessibility validated with deaf users (ASR, TTS, NLP, sign recognition). \u0026bull; High usability metrics: SUS/UEQ in positive ranges, low error rates, clear navigation. \u0026bull; Addresses ethics, privacy, digital equity, and institutional support.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull; Accessibility features are present but only partially or inconsistently implemented. \u0026bull; Partial WCAG compliance (e.g., only Level A) or standards mentioned without evidence. \u0026bull; Moderate usability (average SUS scores, recurring errors). \u0026bull; Multimodal accessibility limited (e.g., only captions or only text).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; Accessibility is addressed only theoretically, without technical implementation. \u0026bull; Does not comply with WCAG or ISO standards. \u0026bull; Offers a single mode of representation (e.g., text only). \u0026bull; No usability metrics or validation with deaf learners.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteractivity (Action and Expression)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull; Provides real bidirectional interaction: sign-language avatars, immediate feedback, multimodal responses. \u0026bull; High feedback precision: high accuracy, low response time, low RMSE. \u0026bull; Applies participation analytics, network analysis, and sentiment analysis. \u0026bull; Intelligent tutoring or feedback systems aligned with learner performance.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull; Interaction exists but is limited to basic inputs (clicks, buttons) with minimal personalization. \u0026bull; Feedback is delayed or generic. \u0026bull; Only basic social interaction available (simple chat or comments). \u0026bull; Limited reporting of feedback metrics (CI, SQ, error rates).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; No functional interactivity. \u0026bull; Bidirectionality is mentioned but not implemented. \u0026bull; No dynamic feedback mechanisms; interaction is static. \u0026bull; No interactivity metrics (e.g., TAM, TAM3, CI).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaptability (Engagement and Motivation)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull; Fully adaptive platforms: real-time adjustment of content, routes, difficulty, and sequencing. \u0026bull; Uses predictive analytics, BKT, cognitive\u0026ndash;affective modeling. \u0026bull; Shows consistent learning gains, low RMSE, and adaptive performance traces. \u0026bull; Adjusts pace and vocabulary dynamically (continual learning). \u0026bull; High retention and sustained personalization with affective-state detection.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull; Adaptation is mentioned but not fully implemented. \u0026bull; Basic adjustments (pace or difficulty) but reactive rather than predictive. \u0026bull; No cognitive\u0026ndash;affective profiling. \u0026bull; Learning gains inconsistent or not reported. \u0026bull; Personalization limited or rule-based.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; No personalization. \u0026bull; Static platform with fixed routes and non-adaptive content. \u0026bull; No adaptive analytics, learner modeling, or dynamic sequencing. \u0026bull; No evaluation using RMSE, BKT, learning gains, or personalization metrics.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"6 Discussion","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe systematic analysis of recent literature reveals that evaluating AI-assisted learning platforms cannot be considered solely from each UDL dimension independently. Rather, their value emerges from the synergistic interaction between accessibility, interactivity, and adaptability, forming a comprehensive educational ecosystem. The convergence of these dimensions allows technology not only to facilitate access but also to transform the learning experience into a dynamic, inclusive, and sustainable process.\u003c/p\u003e \u003cp\u003eRegarding the first research question (how the evaluation of AI platforms has been addressed) it is observed that efforts tend to focus on technical personalization and adaptability. However, a cross-dimensional perspective shows that the effectiveness of adaptability critically depends on accessibility and interactivity: AI models can offer tailored learning paths, but these will only be meaningful if students can perceive, interact with, and receive feedback in comprehensible and contextually appropriate formats (Han et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Sajja et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). This interdependence suggests that measuring each dimension in isolation may underestimate the actual effects on learning and inclusion.\u003c/p\u003e \u003cp\u003eConcerning the second question, which metrics and frameworks are used, the review indicates that normative, perceptual, and algorithmic instruments, although useful, work best when integrated holistically. Usability and user experience metrics (SUS, UEQ) should be linked with adaptive and affective performance indicators, while technical standards (WCAG, ISO) are complemented by real-time learning and interaction data. This integration allows for the assessment of not only technical effectiveness but also the pedagogical and ethical quality of the platform, providing concrete evidence of how AI can facilitate active participation and deep learning.\u003c/p\u003e \u003cp\u003eRegarding the third question, how findings can guide inclusive design based on UDL, the results suggest that the synergy between dimensions offers a reference framework for ethical and pedagogical design decisions. Accessibility ensures entry points, interactivity transforms the experience into meaningful engagement, and adaptability guarantees personalization and sustainability. The integration of cross-dimensional metrics allows the identification of gaps in the student experience, adjustment of pedagogical interventions, and monitoring of algorithmic fairness. In this way, UDL principles go beyond mere regulatory compliance and materialize into effective and equitable learning pathways.\u003c/p\u003e \u003cp\u003eFinally, the discussion highlights tensions and opportunities for future research. While AI can act as a catalyst for inclusion, there remains a need for teacher oversight, data governance, and algorithmic transparency (Kenneth et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Tebourbi et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). True inclusion depends on the platforms\u0026rsquo; ability to coherently integrate accessibility, interactivity, and adaptability, rather than merely on technological sophistication. This perspective also aligns with the Sustainable Development Goals (SDG 4 and 9), suggesting that educational AI can contribute to equity and innovation if conceived within a comprehensive techno-pedagogical framework.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"7 Implications for Future Research","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe findings of this review open several lines of research aimed at strengthening the convergence between UDL and AI in education, from a perspective that integrates technical rigor, pedagogical sense, and ethical commitment.\u003c/p\u003e \u003cp\u003eFirst, there is a clear need to develop comprehensive evaluation models that combine normative metrics, such as ISO/IEC 25022 and WCAG 2.1 guidelines, with pedagogical and affective indicators capable of simultaneously assessing technical accessibility, user experience, and learning outcomes. The literature still lacks standardized frameworks linking software quality to educational quality; advancing empirically validated mixed instruments constitutes a priority challenge for future research.\u003c/p\u003e \u003cp\u003eA second line points to expanding the application contexts of UDL. Most reviewed studies are confined to higher education, leaving basic education scenarios and populations with functional or linguistic diversity underexplored. It will be essential to design and evaluate platforms that incorporate linguistic accessibility (such as automatic sign language translation or gesture recognition) and multimodal adaptations that respond to neurodiversity. Such research can provide evidence on how UDL principles translate into tangible inclusive practices, particularly in learning environments for deaf students or those with visual impairments.\u003c/p\u003e \u003cp\u003eThird, it is necessary to compare the effectiveness of currently employed metrics. Studies rely on usability, engagement, or algorithmic accuracy indicators without establishing their correspondence or cross-validation. Future research should advance toward mixed methodologies that integrate statistical analyses, interaction data mining, and qualitative exploration of the learning experience, so that technical metrics can be interpreted in terms of their pedagogical relevance.\u003c/p\u003e \u003cp\u003eA fourth emerging line involves transparency and algorithmic governance in intelligent educational environments. Few studies examine how to audit AI to detect bias, ensure data privacy, or make personalization processes understandable. Exploring participatory auditing mechanisms, open data policies, or algorithmic equity analyses by student subgroups represents a fertile area for interdisciplinary research.\u003c/p\u003e \u003cp\u003eFurthermore, it will be valuable to investigate in greater depth the relationship between emotional interactivity and motivational adaptability. Integrating affective sensors, multimodal analytics, or neuro educational models would allow researchers to understand how emotions influence personalization effectiveness, paving the way for more empathetic, learner-centered systems.\u003c/p\u003e \u003cp\u003eFinally, future research should reassess the role of teachers in AI educational ecosystems. Results suggest that automation without pedagogical mediation can lead to personalization devoid of educational meaning. Studies on human-AI collaboration is needed to examine how teachers can guide, supervise, and enrich system decisions, ensuring coherence between pedagogical objectives and algorithmic adaptations.\u003c/p\u003e \u003cp\u003eMore broadly, this review calls for an epistemological shift: moving from research focused on technical efficiency toward a deeper understanding of comprehensive, inclusive, and ethical evaluation of AI in education. Advancing evidence on how accessibility, interactivity, and adaptability metrics translate into genuine learning, well-being, and equity will be the next necessary step to consolidate a new field of study: universal evaluation of AI-assisted learning.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eConflict of interest\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no known financial or non-financial competing interests that could have appeared to influence the work reported in this paper.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThe authors declare that no funds, grants, or other financial support were received during the preparation of this manuscript.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eFrank Alexander Parra conceived the study, defined the research objectives, and led the design of the systematic review. He coordinated the development of the analytical framework based on Universal Design for Learning (UDL) and supervised the construction of the evaluation metrics and rubrics presented in the study. Juan Pablo Rivera contributed to the literature search strategy, screening process, data extraction, and qualitative synthesis of the reviewed studies. He also participated in the analysis of evaluation metrics and in the drafting and revision of the manuscript. Both authors independently screened the retrieved studies in multiple rounds (title and abstract, full text, and methodological appraisal) following the inclusion and exclusion criteria. Disagreements were resolved through discussion and consensus. Inter-coder reliability was calculated using Cohen\u0026rsquo;s κ (κ = 0.84), indicating substantial agreement. The academic advisors provided methodological guidance, critical feedback on the analytical framework, and reviewed the manuscript for conceptual coherence and academic rigor. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated and analyzed during the current study, including the coding matrix and extracted data from the reviewed articles, are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAdomavicius, G., Tuzhilin, A.: Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-Art and Possible Extensions. IEEE Trans. Knowl. Data Eng. \u003cb\u003e17\u003c/b\u003e(6), 734\u0026ndash;749 (2005). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TKDE.2005.99\u003c/span\u003e\u003cspan address=\"10.1109/TKDE.2005.99\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAra, J., SikLanyi, C., Kelemen, A.: Accessibility engineering in web evaluation process: a systematic literature review. Univ. Access Inf. Soc. \u003cb\u003e23\u003c/b\u003e(2), 653\u0026ndash;686 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10209-023-00967-2\u003c/span\u003e\u003cspan address=\"10.1007/s10209-023-00967-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarbu, M., Iordache, D.D., Petre, I., Barbu, D.C., Băjenaru, L.: Framework Design for Reinforcing the Potential of XR Technologies in Transforming Inclusive Education [Article]. Appl. Sci. (Switzerland). \u003cb\u003e15\u003c/b\u003e(3) (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003eArticle 1484. https://doi.org/10.3390/app15031484\u003c/span\u003e\u003cspan address=\"Article 1484. 10.3390/app15031484\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarkovska, O., Ruban, I., Tymoshenko, D., Holovchenko, O., Yankovskyi, O., RESEARCH ON MOBILE MACHINE LEARNING PLATFORMS FOR HUMAN GESTURE RECOGNITION IN HUMAN-MACHINE INTERACTION SYSTEMS [Article]: Technol. Audit Prod. Reserves. \u003cb\u003e2\u003c/b\u003e(82), 6\u0026ndash;14 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.15587/2706-5448.2025.325423\u003c/span\u003e\u003cspan address=\"10.15587/2706-5448.2025.325423\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarsalou, L.W.: Language comprehension: Archival memory or preparation for situated action. Discourse Processes. \u003cb\u003e28\u003c/b\u003e(1), 61\u0026ndash;80 (1999)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBercaru, V., Popescu, N.: A Systematic Review of Accessibility Techniques for Online Platforms: Current Trends and Challenges. \u003cem\u003eApplied Sciences\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(22). (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBotella, P., Burgu\u0026eacute;s, X., Carvallo, J., Franch, X., Grau, G., Marco, J., Quer, C.: ISO/IEC 9126 in practice: what do we need to know? (2004)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCAST: Universal Design for Learning Guidelines version 2.2. In. (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDemartini, C.G., Sciascia, L., Bosso, A., Manuri, F.: Artificial Intelligence Bringing Improvements to Adaptive Learning in Education: A Case Study. Sustainability, \u003cb\u003e16\u003c/b\u003e(3). (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEl-Sabagh, H.A., Hamed, S.M.: Adaptive E-learning Systems Based on Learning Styles: A UDL Perspective. Int. J. Educational Technol. High. Educ. \u003cb\u003e18\u003c/b\u003e, 1\u0026ndash;22 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s41239-021-00289-4\u003c/span\u003e\u003cspan address=\"10.1186/s41239-021-00289-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGee, J.P.: Reading as Situated Language: A Sociocognitive Perspective. J. Adolesc. Adult Lit. \u003cb\u003e44\u003c/b\u003e(8), 714\u0026ndash;725 (2001)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHalkiopoulos, C., Gkintoni, E.: Leveraging AI in E-Learning: Personalized Learning and Adaptive Assessment through Cognitive Neuropsychology\u0026mdash;A Systematic Analysis. \u003cem\u003eElectronics\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(18). (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHan, J., Liu, G., Liu, X.: Continue using or gathering dust? A mixed method research on the factors influencing the continuous use intention for an AI-powered adaptive learning system for rural middle school students. Heliyon. \u003cb\u003e10\u003c/b\u003e(12), e33251 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.heliyon.2024.e33251\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2024.e33251\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHerrera Nieves, L., Moya, C., E., Montes Soldado, R., ASSESSMENT METHODOLOGY USING THE UNIVERSAL DESIGN FOR LEARNING PERSPECTIVE: Turkish Online J. Distance Educ. \u003cb\u003e26\u003c/b\u003e(3), 238\u0026ndash;255 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.17718/tojde.1510242\u003c/span\u003e\u003cspan address=\"10.17718/tojde.1510242\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e MOODLE USABILITY\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHerrera Nieves, L.B., Moya, C., E., Soldado, M., R: Moodle Usability Assessment Methodology using the Universal Design for Learning perspective. In: Anadolu University (2025)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHssina, B., Erritali, M.: A Personalized Pedagogical Objectives Based on a Genetic Algorithm in an Adaptive Learning System. Procedia Comput. Sci. \u003cb\u003e151\u003c/b\u003e, 1152\u0026ndash;1157 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.procs.2019.04.164\u003c/span\u003e\u003cspan address=\"10.1016/j.procs.2019.04.164\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIfenthaler, D., Gibson, D.: Adoption of Data Analytics in Higher Education Learning and Teaching. Springer (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-030-48884-0\u003c/span\u003e\u003cspan address=\"10.1007/978-3-030-48884-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIngav\u0026eacute;lez-Guerra, P., Ot\u0026oacute;n-Tortosa, S., Hilera-Gonz\u0026aacute;lez, J., S\u0026aacute;nchez-Gord\u0026oacute;n, M.: The use of accessibility metadata in e-learning environments: a systematic literature review. Univ. Access Inf. Soc. \u003cb\u003e22\u003c/b\u003e(2), 445\u0026ndash;461 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10209-021-00851-x\u003c/span\u003e\u003cspan address=\"10.1007/s10209-021-00851-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIsaeva, R., Karasartova, N., Dznunusnalieva, K., Mirzoeva, K., Mokliuk, M., THROUGH ADAPTIVE LEARNING PLATFORMS AND EMERGING COMPUTER TECHNOLOGIES IN EDUCATION: Jurnal Ilmiah Ilmu Terapan Universitas Jambi. \u003cb\u003e9\u003c/b\u003e(1), 144\u0026ndash;160 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.22437/jiituj.v9i1.37967\u003c/span\u003e\u003cspan address=\"10.22437/jiituj.v9i1.37967\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e ENHANCING LEARNING EFFECTIVENESS\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIsmailova, R., Inal, Y.: Comparison of Online Accessibility Evaluation Tools: An Analysis of Tool Effectiveness. IEEE Access. \u003cb\u003e10\u003c/b\u003e, 58233\u0026ndash;58239 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ACCESS.2022.3179375\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2022.3179375\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eISO: \u003cem\u003eISO 9241\u0026thinsp;\u0026ndash;\u0026thinsp;210:2019 Ergonomics of human-system interaction \u0026mdash; Part 210: Human-centred design for interactive systems\u003c/em\u003e. (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.iso.org/standard/77520.html\u003c/span\u003e\u003cspan address=\"https://www.iso.org/standard/77520.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eISO, \u0026amp; IEC: ISO/IEC 25022: Systems and Software Engineering \u0026mdash; Systems and Software Quality Requirements and Evaluation (SQuaRE) \u0026mdash; Measurement of Quality in Use. ISO/IEC, In. Geneva, Switzerland (2016)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKenneth, H., Marino, M., Vasquez, E., Taub, M., Hunt, J., Tazi, Y.: Navigating AI-Powered Personalized Learning in Special Education: A Guide for Preservice Teacher Faculty. J. Special Educ. Preparation. \u003cb\u003e4\u003c/b\u003e(2), 90\u0026ndash;95 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.33043/5b2xqcb3\u003c/span\u003e\u003cspan address=\"10.33043/5b2xqcb3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhosravi, H., Denny, P., Moore, S., Stamper, J.: Learnersourcing in the age of AI: Student, educator and machine partnerships for content creation. Computers Education: Artif. Intell. \u003cb\u003e5\u003c/b\u003e, 100151 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.caeai.2023.100151\u003c/span\u003e\u003cspan address=\"10.1016/j.caeai.2023.100151\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLandis, J.R., Koch, G.G.: The Measurement of Observer Agreement for Categorical Data. Biometrics. \u003cb\u003e33\u003c/b\u003e(1), 159\u0026ndash;174 (1977). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2307/2529310\u003c/span\u003e\u003cspan address=\"10.2307/2529310\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLu, J.: Artificial Intelligence in Education Design and Evaluation of Adaptive Learning Systems. Region - Educational Res. Reviews. \u003cb\u003e6\u003c/b\u003e(4), 209\u0026ndash;214 (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMadahana, M., Khoza-Shangase, K., Moroe, N., Mayombo, D., Nyandoro, O., Ekoru, J.: A proposed artificial intelligence-based real-time speech-to-text to sign language translator for South African official languages for the COVID-19 era and beyond: In pursuit of solutions for the hearing impaired [Article]. South Afr. J. Commun. Disord. \u003cb\u003e69\u003c/b\u003e(2) (2022). Article a915 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4102/sajcd.v69i2.915\u003c/span\u003e\u003cspan address=\"10.4102/sajcd.v69i2.915\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMadanchian, M., Taherdoost, H.: Examining Critical Factors in Selecting AI Tools for Educational Success. Procedia Comput. Sci. \u003cb\u003e263\u003c/b\u003e, 923\u0026ndash;933 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.procs.2025.07.111\u003c/span\u003e\u003cspan address=\"10.1016/j.procs.2025.07.111\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaksymov, M., Davydov, V., Petik, T., Grishyn, M.: Development of an Assessment Methodology for Educational Platforms in the Context of Educators\u0026rsquo; Innovation and Research Activities. Filosofiya osvity Philos. Educ. \u003cb\u003e31\u003c/b\u003e, 90\u0026ndash;110 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.31874/2309-1606-2025-31-1-6\u003c/span\u003e\u003cspan address=\"10.31874/2309-1606-2025-31-1-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMehmood, H.: Artificial intelligence in educational governance: Implications for adaptive learning and student mindset formation. Rev. Educ. Adm. Law. \u003cb\u003e8\u003c/b\u003e(2), 343\u0026ndash;359 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.47067/real.v8i2.429\u003c/span\u003e\u003cspan address=\"10.47067/real.v8i2.429\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMemari, M., Taheri, A.: Adaptive Teaching of the Iranian Sign Language Based on Continual Learning Algorithms. IEEE Access. \u003cb\u003e12\u003c/b\u003e, 164164\u0026ndash;164177 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ACCESS.2024.3492056\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2024.3492056\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMontes, A., Jim\u0026eacute;nez, R., Garc\u0026iacute;a, F.: Evaluating Moodle Usability from a UDL Perspective: A Case Study in Latin American Higher Education. Educ. Inform. Technol. \u003cb\u003e29\u003c/b\u003e(3), 455\u0026ndash;478 (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorgado, E., Leonido, L., Pereira, A., Gouveia, L.B.: Technology-Mediated Education: impact of AI on the main distance learning modalities [Article]. Educational Process: Int. J. \u003cb\u003e16\u003c/b\u003e (2025). Article e2025211 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.22521/edupij.2025.16.211\u003c/span\u003e\u003cspan address=\"10.22521/edupij.2025.16.211\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eObeid, A., Ibrahim, R., Fadhil, A.: Extended Model of Expectation Confirmation Model to Examine Users\u0026rsquo; Continuous Intention Toward the Utilization of E-Learning Platforms. IEEE Access. \u003cb\u003e12\u003c/b\u003e, 40752\u0026ndash;40764 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ACCESS.2024.3373190\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2024.3373190\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePage, M.J., McKenzie, J.E., Bossuyt, P.M., Boutron, I., Hoffmann, T.C., Mulrow, C.D., Moher, D.: The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. \u003cb\u003e372\u003c/b\u003e, n71 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmj.n71\u003c/span\u003e\u003cspan address=\"10.1136/bmj.n71\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParody, L.M., Leiva, J.-J., Santos-Villalba, M.J.: El Dise\u0026ntilde;o Universal para el Aprendizaje en la Formaci\u0026oacute;n Digital del Profesorado desde una Mirada Pedag\u0026oacute;gica Inclusiva. Revista Latinoam. de educaci\u0026oacute;n inclusiva. \u003cb\u003e16\u003c/b\u003e, 109\u0026ndash;123 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4067/S0718-73782022000200109\u003c/span\u003e\u003cspan address=\"10.4067/S0718-73782022000200109\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParveen, K., Obidallah, W.J., Alghamdi, A.A., Alduraywish, Y.A., Shafiq, M.: Gesture-enhanced adaptive learning platform for personalized AI-driven education. Interact. Learn. Environ., 1\u0026ndash;22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10494820.2025.2542897\u003c/span\u003e\u003cspan address=\"10.1080/10494820.2025.2542897\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePunitham, M., Balasubramaniam, P., Lakshmi, S., Sriram, N., Shivakumar, K.M.: Navigating open and distance learning trends, advantages, challenges and innovations for the future education - A systematic review. Plant. Sci. Today. \u003cb\u003e12\u003c/b\u003e(2) (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.14719/pst.8108\u003c/span\u003e\u003cspan address=\"10.14719/pst.8108\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajabi, K.I.: Beyond the LMS: How a Multi-Platform, AI-Enhanced Ecosystem Boosts Motivation, Interaction, Deep Learning, and Design Competence among Arabic-Speaking Graduate Students. (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21203/rs.3.rs-7156104/v1\u003c/span\u003e\u003cspan address=\"10.21203/rs.3.rs-7156104/v1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRodr\u0026iacute;guez-Moreno, I., Mart\u0026iacute;nez-Otzeta, J.M., Sierra, B.: HAKA: HierArchical Knowledge Acquisition in a sign language tutor. Expert Syst. Appl. \u003cb\u003e215\u003c/b\u003e, 119365 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.eswa.2022.119365\u003c/span\u003e\u003cspan address=\"10.1016/j.eswa.2022.119365\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSabor\u0026iacute;o-Taylor, A., Rojas-Ram\u0026iacute;rez, G.: Universal design for learning and artificial intelligence in the digital era: Fostering inclusion and autonomous learning. Int. J. Prof. Dev. Learners Learn. \u003cb\u003e6\u003c/b\u003e(2), ep2408 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.30935/ijpdll/14694\u003c/span\u003e\u003cspan address=\"10.30935/ijpdll/14694\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSabor\u0026iacute;o-Taylor, M., Rojas-Ram\u0026iacute;rez, J.: Universal design for learning and artificial intelligence in the digital era: Fostering inclusion and autonomous learning. Int. J. Prof. Dev. Learners Learn. \u003cb\u003e6\u003c/b\u003e(2), ep2408 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.30935/ijpdll/14694\u003c/span\u003e\u003cspan address=\"10.30935/ijpdll/14694\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSajja, R., Sermet, Y., Cwiertny, D., Demir, I.: Integrating AI and Learning Analytics for Data-Driven Pedagogical Decisions and Personalized Interventions in Education. \u003cem\u003eTechnology, Knowledge and Learning\u003c/em\u003e. (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10758-025-09897-9\u003c/span\u003e\u003cspan address=\"10.1007/s10758-025-09897-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSheejamol, P.T., Chacko, A.M., Kumar, S.D.M.: Beyond the One-Size-Fits-All: A Systematic Review of Personalized and Gamified e-Learning for Neurodivergent Learners [Article]. Electron. J. e-Learning. \u003cb\u003e23\u003c/b\u003e(3), 101\u0026ndash;119 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.34190/ejel.23.3.4051\u003c/span\u003e\u003cspan address=\"10.34190/ejel.23.3.4051\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSlepankova, M., Kilianova, K., Kockova, P., Kostolanyova, K., Kotyrba, M., Habiballa, H.: Student Perceptions and Preferences in Personalized AI-driven Learning [Article]. Acta Informatica Pragensia. \u003cb\u003e14\u003c/b\u003e(2), 261\u0026ndash;271 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18267/j.aip.278\u003c/span\u003e\u003cspan address=\"10.18267/j.aip.278\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTan, L.Y., Hu, S., Yeo, D.J., Cheong, K.H.: Artificial intelligence-enabled adaptive learning platforms: A review. Computers Education: Artif. Intell. \u003cb\u003e9\u003c/b\u003e, 100429 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.caeai.2025.100429\u003c/span\u003e\u003cspan address=\"10.1016/j.caeai.2025.100429\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTebourbi, H., Nouzri, S., Mualla, Y., Abbas-Turki, A.: Artificial Intelligence Agents for Personalized Adaptive Learning. Procedia Comput. Sci. \u003cb\u003e265\u003c/b\u003e, 252\u0026ndash;259 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.procs.2025.07.179\u003c/span\u003e\u003cspan address=\"10.1016/j.procs.2025.07.179\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUNESCO: \u003cem\u003eConsenso de Beijing sobre la inteligencia artificial y la educaci\u0026oacute;n\u003c/em\u003e. (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://unesdoc.unesco.org/ark:/48223/pf0000368303\u003c/span\u003e\u003cspan address=\"https://unesdoc.unesco.org/ark:/48223/pf0000368303\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVanLehn, K.: The Behavior of Tutoring Systems. Int. J. Artif. Intell. Educ. \u003cb\u003e16\u003c/b\u003e(3), 227\u0026ndash;265 (2006)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWangdi, T., Shimray, R.: AI-Powered ReadTheory as a Self-Access Learning Platform to Enhance EFL Learners\u0026rsquo; Reading Enjoyment and Comprehension Skills: A Posthumanist Perspective [Article]. SiSal J. \u003cb\u003e16\u003c/b\u003e(2), 437\u0026ndash;460 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.37237/160209\u003c/span\u003e\u003cspan address=\"10.37237/160209\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZapata, C., Acosta, J.: Innovaciones Tecnol\u0026oacute;gicas Para Inclusi\u0026oacute;n Educativa De Estudiantes Sordos. Ingenier\u0026iacute;a e Innovaci\u0026oacute;n. \u003cb\u003e6\u003c/b\u003e(2) (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21897/23460466.1595\u003c/span\u003e\u003cspan address=\"10.21897/23460466.1595\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Y., Han, Y., Zhu, Z., Jiang, X.: Artificial intelligence in sign language recognition: A comprehensive bibliometric and visual analysis [Article]. Comput. Electr. Eng. \u003cb\u003e120\u003c/b\u003e, 109854 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compeleceng.2024.109854\u003c/span\u003e\u003cspan address=\"10.1016/j.compeleceng.2024.109854\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence, Universal Design for Learning, Inclusive Learning Environments, Accessibility, Interactivity, Evaluation Metrics","lastPublishedDoi":"10.21203/rs.3.rs-8801285/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8801285/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial Intelligence is reshaping the design, evaluation, and personalization of digital learning environments, enabling adaptive and data-driven pedagogies that respond to diverse learner needs. In parallel, the Universal Design for Learning (UDL) framework has become central to inclusive education, offering principles to ensure accessibility, engagement, and multiple means of representation. Despite this convergence, systematic analyses that evaluate AI-based learning platforms through UDL remain scarce.\u003c/p\u003e \u003cp\u003eExisting reviews on AI in education have primarily focused on algorithmic efficiency, adaptive architectures, or technological innovation. However, they lack an analytical framework that connects AI-based learning technologies with UDL principles, largely because they do not articulate the dimensions needed to operationalize these principles into evaluative criteria. As a result, most AI-driven platforms are assessed in terms of technical performance rather than pedagogical inclusiveness, usability, accessibility compliance, or learner engagement. To address this gap, this study conducts a systematic review of evaluation metrics applied to AI-based learning platforms, following the PRISMA methodology and analyzing peer-reviewed studies published from 2019 onward. Using UDL as a conceptual and analytical scaffold, the review structures its synthesis around three operational dimensions derived from the framework: accessibility (representation), interactivity (action and expression), and adaptability (engagement and motivation).\u003c/p\u003e \u003cp\u003eBuilding on this analytical approach, the purpose of the review is twofold: first, to map the state of the art in evaluating AI-driven learning platforms through both normative and algorithmic metrics; and second, to propose an integrative model that links international standards with user-experience indicators and adaptive performance measures. In doing so, the study contributes a structured evaluative perspective that bridges technical methodologies and pedagogical frameworks for inclusion, advancing the development of more equitable, transparent, and accessible AI-based learning systems.\u003c/p\u003e","manuscriptTitle":"AI-Based Learning Platforms: A Systematic Review of Evaluation Metrics for Accessibility, Interactivity and Adaptability through the Lens of Universal Design for Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-27 03:02:08","doi":"10.21203/rs.3.rs-8801285/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"90705f65-7ca6-4fe5-b304-d71167115bde","owner":[],"postedDate":"February 27th, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Rejected","date":"2026-05-14T16:48:19+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-12T09:11:34+00:00","index":18,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-09T13:38:25+00:00","index":17,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-29T13:02:59+00:00","index":16,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-14T16:54:33+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-27 03:02:08","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8801285","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8801285","identity":"rs-8801285","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}