Conceptualizing and Validating the 4R Model of Generative AI-Based Emotional Support Scale: Regulation, Resonance, Reinforcement, and Reflection

Conceptualizing and Validating the 4R Model of Generative AI-Based Emotional Support Scale: Regulation, Resonance, Reinforcement, and Reflection | 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 Article Conceptualizing and Validating the 4R Model of Generative AI-Based Emotional Support Scale: Regulation, Resonance, Reinforcement, and Reflection Yanchao Yang, Xinxin Yang, Zehang Tan, Qiu Wei, Xiaofeng Wang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6956402/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 This study conceptualizes and validates the 4R Model of Generative AI-Based Emotional Support Scale, encompassing four key dimensions: Regulation (5 items), Resonance (6 items), Reinforcement (5 items), and Reflection (4 items). These dimensions were developed based on expert interviews and are intended to assess the effectiveness of generative AI in providing emotional support. The study sample consisted of 996 participants from a northern Chinese university, with a convenience sampling method used to collect data. The validation process followed multiple steps, including item analysis, exploratory factor analysis, and confirmatory factor analysis, assessing its reliability, content validity, structural validity, convergent validity and discriminant validity. The findings indicate that the scale demonstrates satisfactory reliability and construct validity. However, while overall construct validity is acceptable, the scale exhibited some challenges in discriminant validity. This limitation can be attributed to several factors, including overlapping contextual assumptions, the interwoven nature of different support types, the lack of diversity in the sample, and the overlap of multiple response strategies. These factors likely contributed to the observed difficulties in achieving strong discriminant validity. Key limitations of the study include concerns regarding the representativeness of the sample, the cross-sectional design that precluded the assessment of test-retest reliability, the reliance on a single validation method, and the focus on generative AI technology in general rather than specific AI-based products. Corresponding solutions are proposed, including expanding the sample to ensure greater diversity, employing a longitudinal design to assess test-retest reliability, incorporating multiple validation methods, and focusing on specific AI-based products in future research. In conclusion, while the 4R Model offers a robust framework for evaluating generative AI-based emotional support, further research is necessary to refine the scale, particularly in terms of improving discriminant validity and expanding its applicability across more diverse contexts. Physical sciences/Mathematics and computing/Information technology Biological sciences/Psychology Generative Artificial Intelligence Emotional Support Regulation Resonance Reinforcement Reflection Figures Figure 1 1. Introduction Mental health, referring to a state of emotional and psychological well-being that allows individuals to manage life's challenges, recognize their strengths, perform effectively in learning and work environments, and positively contribute to their communities (WHO, 2022), has consistently been a research focus in academic fields (X.-Y. Li, 2025). Many studies have delved into the antecedents of mental health issues, investigating a range of factors such as social and economic conditions, including poverty (Díaz et al., 2022; Thomson et al., 2023; Zaneva et al., 2022), violence (Polanin et al., 2021; Rashid Soron et al., 2021), and inequality (Gibson et al., 2021; W. Li et al., 2021; Tibber et al., 2022), and adverse experiences during critical periods. Concurrently, a substantial body of research has explored the consequences of mental health on various domains of life, including academic performance (Agnafors et al., 2021; Chu et al., 2023; Duncan et al., 2021; Hinkson Jr et al., 2022), wellbeing (O’Connor et al., 2021; Rider et al., 2021) and organizational performance (Gilbert et al., 2024), demonstrating its profound influence on individuals' capacity to function and thrive in both personal and professional contexts. One important area of research in mental health is exploring how to improve mental health through social support (Ault et al., 2021; Bedaso et al., 2021; Y. Huang et al., 2021; Zhou & Cheng, 2022). Social support is a multifaceted concept that has garnered significant attention across various fields of research, particularly in psychology, education, and healthcare. It generally refers to the assistance and resources provided by others—whether individuals, groups, or organizations—that help individuals manage stress (Guo et al., 2021; Karadaş & Duran, 2022), foster resilience (Mai et al., 2021; Permatasari et al., 2021), and enhance their well-being (Holliman et al., 2021; L. Huang & Zhang, 2022; Shuo et al., 2022) and learning engagement (Barratt & Duran, 2021). Perceived social support can be divided into four key dimensions: emotional support, informational support, instrumental support, and appraisal support (House, 1983). Among these, emotional support stands out as a crucial factor in mental health, as it involves providing empathy, care, and understanding during times of distress. Emotional support from family, friends, and even colleagues can significantly reduce the impact of stress, increase feelings of security, and enhance emotional resilience. The presence of emotional support, whether in family settings or organizational contexts, has been shown to buffer individuals against mental health challenges, facilitating better overall well-being and functioning (Buck et al., 2024; X. Hu et al., 2022; Zeng, 2024). Currently, emotional support is predominantly provided by individuals such as teachers (Kikas & Tang, 2019; Schenke et al., 2018; Yeung & Leadbeater, 2010) and parents (Gaspar et al., 2022; Romm et al., 2021). However, individuals experiencing mental health challenges often face the fear of being stigmatized or labeled (Goel et al., 2023; Jauch et al., 2023; Steiger et al., 2022), which can act as a significant barrier to seeking traditional forms of support. This stigma can lead to reluctance in disclosing one's struggles, thus exacerbating feelings of isolation and distress. Consequently, the need for confidential and non-judgmental emotional support has become increasingly critical for individuals facing mental health issues. Human-chatbot interactions have indeed evolved significantly over time. The emergence of AI-generated content (AIGC) technologies presents new opportunities to address this gap, offering individuals the ability to access emotional support in a private and anonymous manner (Arifudin et al., 2024; Mendoza-Pinto, 2023), thanks to affective computing (Picard, 2000). Affective computing can perform (a) emotion recognition (Cheng Lin et al., 2013), which involves text sentiment analysis, speech emotion recognition, facial expression recognition, and physiological signal processing; (b) emotion understanding (Moore, 2017; Wang et al., 2022), which focuses on developing context-aware systems; and (c) emotion generation (Pei et al., 2024), which includes emotional speech synthesis, therapeutic music generation, emotional visual content creation, and dialogue generation systems. Recent empirical research has begun to investigate the role of AIGC in delivering personalized, confidential emotional support, with promising findings suggesting that it can facilitate improved mental well-being and mitigate the stigma associated with seeking help (Etori & Gini, 2024; Herbener & Damholdt, 2025; Naher, 2024; Shin et al., 2022). These studies underscore the potential of AIGC to serve as a valuable resource for enhancing emotional resilience and promoting mental health, particularly in environments where traditional support systems may not be readily accessible or where privacy is a key concern. Generative Artificial Intelligence has been widely applied in emotional support scenarios, offering anonymous and personalized emotional care. However, AIGC itself is often regarded as a "black-box" system (Bearman & Ajjawi, 2023; Hassija et al., 2024; Rohan et al., 2023), meaning that its inner workings and decision-making processes are not easily understood by external observers. Therefore, when further improving these models to effectively provide emotional support, it becomes crucial to accurately differentiate between various types of emotional support. Thus, this paper primarily constructs and validates a multidimensional emotional support model to optimize the application of AIGC in emotional support. Through this model, we can identify and distinguish the different dimensions of emotional support, such as emotional resonance, emotional regulation, non-judgmental listening, and emotional enhancement. This provides clear guidance for AIGC systems, ensuring that their responses can more precisely align with users' emotional needs. 2. Methods 2.1 Ethical considerations Prior to data collection, ethical approval was obtained from the Research Ethics Committee of the corresponding author's institution (Approval No. MMCIRB-2024-001). The application outlined the study’s aims, methods, privacy protections, and participant criteria, and confirmed that no identifiable information (e.g., names or student IDs) would be collected. Data were gathered via convenience sampling at an independent college in northern China. The survey link, which included an electronic informed consent form, was distributed through instructors via WeChat. Participation was voluntary, and students had to click "Agree to Participate" to proceed. All data were anonymized and securely stored for research use only. All methods were performed in accordance with the Declaration of Helsinki. 2.2 Participants Data collection took place in December 2024, with each participant requiring approximately 5 minutes to complete the survey. To ensure data quality, the collected data were filtered based on response time and Mahalanobis distance, resulting in 996 valid participants. The sample was then randomly divided into two groups for statistical analysis. As shown in Table 1 , the first group, used for exploratory factor analysis (EFA), consisted of 496 participants, including 179 males (36.10%) and 317 females (63.90%), with 103 participants (20.80%) from urban areas and 393 (79.20%) from rural areas. The second group, used for confirmatory factor analysis (CFA), included 500 participants, comprising 174 males (34.80%) and 326 females (65.20%), with 117 participants (23.40%) from urban areas and 383 (76.60%) from rural areas. Table 1 Demographic information EFA CFA Variable Group N % N % Faculty FEEE 61 12.30% 70 14.00% FE 127 25.60% 122 24.40% FIL 93 18.80% 82 16.40% FB 30 6.00% 30 6.00% FPE 22 4.40% 20 4.00% FFTAD 163 32.90% 176 35.20% Sex Male 179 36.10% 174 34.80% Female 317 63.90% 326 65.20% Birthplace Urban 103 20.80% 117 23.40% Rural 393 79.20% 383 76.60% Note: FEEE = Faculty of Electrical and Information Engineering, FE = Faculty of Engineering, FIL = Faculty of International Languages, FB = Faculty of Business, FPE = Faculty of Physical Education, FFTAD = Faculty of Film, Television, and Art Design 2.3 Instruments 2.3.1 4R Model of Generative AI-Based Emotional Support Scale The 4R Model of Generative AI-Based Emotional Support was developed to explore how generative AI can provide effective emotional support. The model is based on expert interviews, qualitative analysis, and rigorous validation. First, five experts were interviewed, including one IT specialist, the former director of a big data center at a Fortune 500 company in China with expertise in AI, an educational measurement expert with multiple SSCI-indexed publications on scale development, an assistant professor with research interest on healing, and a psychology doctoral student. The protocol was developed through an iterative process involving a review of existing literature on emotional support, generative AI interaction, and human-computer communication. The aim was to elicit expert insights on how generative AI can provide emotional support in realistic user-AI interactions. Sample questions were like “When users seek emotional or psychological help, how does Gen-AI typically respond”. Second, the interview for each one lasted approximately 30 minutes, and the interview protocol focused on the potential of generative AI to support emotional well-being, emphasizing its mechanisms and practical applications. Third, the interview recordings were first transcribed, ensuring that every response was captured accurately. Once transcriptions were completed, the data was analyzed using MAXQDA software, which allowed for in-depth qualitative analysis. Following grounded theory methodology, the analysis was conducted in three stages. To begin with, during initial coding, the transcripts were reviewed line by line, with open coding applied to identify key themes and concepts related to emotional support. Each relevant statement was assigned a preliminary code. Next, in the axial coding stage, the initial codes were grouped into categories based on similarities and relationships, refining the themes related to how generative AI can provide emotional support. Finally, during selective coding, the most significant categories were identified and related to the central research question, leading to the organization and definition of four core categories: Regulation (5 items), Resonance (6 items), Reinforcement (5 items), and Reflection (4 items). Specifically, Regulation refers to AI's ability to help users manage their emotions by offering tools and strategies to cope with stress or negative feelings, such as relaxation techniques One sample item is “Generative artificial intelligence guides me to express my emotions, helping me release them through conversation or other means.”. Resonance involves AI creating an emotional connection by understanding and responding to users' emotions, making them feel heard and validated. One sample item is “When I confide in generative artificial intelligence, it always maintains an open mindset and does not judge my emotions or experiences.” Reinforcement is AI’s provision of positive feedback and encouragement, which boosts users' confidence and motivates them to continue positive behaviors. One sample item is ” Whenever I achieve a small accomplishment, generative artificial intelligence acknowledges my efforts and uses positive words to boost my confidence.” Finally, Reflection emphasizes AI's role in prompting users to reflect on their emotions and behaviors, aiding them in gaining deeper insights into their feelings and reactions for personal growth. One sample item is “Generative artificial intelligence helps me review my emotions and analyze what events or situations triggered them, so I can understand why I have these feelings.” Fourth, this model was further validated through expert evaluations, where eight university counselors with backgrounds in psychology, education, and student mental health counseling assessed the scale items for clarity, representativeness, and relevance. The item-level content validity index (I-CVI) and scale-level content validity index (S-CVI) both exceeded 0.9, confirming the scale's strong validity. The final scale adopted a five-point Likert format, where higher scores indicated stronger agreement with the described dimensions, ensuring alignment with user experiences and interpretability. 2.4 Analytical procedure The scale development process began with a comprehensive item analysis to assess the discriminatory power of each scale item. Participants were divided into two groups: the top 27% and the bottom 27%, based on their overall scores, enabling a comparison of how well the items differentiated individuals with high versus low levels of the targeted construct. To analyze these differences, an independent samples t-test was conducted for each item. The t-test allowed for comparison of the mean scores between the two groups, with the 95% confidence interval (CI) being used to assess the precision of the estimated difference in means. Items that demonstrated significant differences between the high and low groups, as indicated by a p -value less than 0.05 and a non-overlapping 95% CI, were retained for the final scale. Conversely, items that did not exhibit significant differences (non-significant t-test results and overlapping 95% CI) were excluded, ensuring that only those items with strong discriminatory power were incorporated into the scale. The next step in the scale development process involved item-total correlation analysis, which served as a basis for evaluating the homogeneity of individual items. In this analysis, the correlation between each item and the total score was computed. Items that exhibited lower correlations with the total score were considered to have lower homogeneity with the overall construct and were flagged for potential removal. A threshold of 0.4 was set for the item-total correlation (Loiacono et al., 2002; Wolfinbarger & Gilly, 2003); items with a correlation below this value were deemed to lack sufficient homogeneity and were deleted from the scale. Conversely, items with higher correlations were retained, as they demonstrated strong alignment with the overall construct being measured. The analysis of "Cronbach's α if item deleted" was also conducted to further assess each item's contribution to the internal consistency of the scale. If deleting an item significantly increased the Cronbach's α, it indicated the item contributed little to consistency and could be removed. Conversely, if the Cronbach's α decreased upon removal, the item was essential for the scale's internal consistency and should be retained. The Kaiser-Meyer-Olkin (KMO) measure and Bartlett’s Test of Sphericity were used via JASP to assess whether the data were suitable for factor analysis. A KMO value closer to 1 indicated that the data were suitable for factor analysis, with a value greater than 0.8 generally considered acceptable (Kaiser, 1974). If the KMO value was low, it suggested that the correlations between variables were insufficient for factor analysis. Bartlett’s Test of Sphericity checked whether the correlation matrix was an identity matrix. A significant result ( p -value < 0.05) indicated that there were sufficient correlations among the variables, making it appropriate to proceed with factor analysis. Following this, exploratory factor analysis (EFA) was performed via JASP to identify the latent factor structure underlying the scale. EFA revealed the key factors that explained the most variance in the data, providing insight into how items clustered into coherent dimensions. Once the factor structure was established, confirmatory factor analysis (CFA) was conducted to validate the model with SmartPLS (CB-SEM module). A series of fit indices, including RMSEA, SRMR, CFI, TLI, and χ2 statistics with degrees of freedom ( df ), were used to assess model fit. A well-fitting model was characterized by RMSEA and SRMR values below 0.08, and CFI and TLI values above 0.90 (Hu & Bentler, 1999; McDonald & Ho, 2002). Finally, the scale’s construct validity was rigorously evaluated through convergent and discriminant validity analyses. To assess convergent validity, three key indicators were used: factor loadings, composite reliability (CR), and average variance extracted (AVE) (Fornell & Larcker, 1981; Hair et al., 2010). First, factor loadings were examined. These loadings represent the correlation between each item and its associated factor. A high factor loading (ideally greater than 0.7) suggests that the item is strongly related to the underlying construct it is meant to measure. Items with lower factor loadings were considered less relevant and were potentially removed to ensure the scale measures the intended concept accurately. Second, CR was calculated. CR evaluates the internal consistency of a construct, which refers to how reliably the items within that construct measure the same underlying idea. A CR value above 0.7 indicates acceptable internal consistency, suggesting that the items are cohesive in their measurement of the construct. Finally, AVE was assessed to determine how much variance in the items is explained by the factor, compared to the error variance. An AVE value of 0.5 or higher indicates that more than half of the variance in the items is explained by the construct, thus confirming that the items are valid indicators of the construct. Discriminant validity refers to the extent to which a construct is truly distinct from other constructs in the model, ensuring that it does not overlap significantly with other constructs. One common method for assessing discriminant validity is the Fornell-Larcker criterion (Fornell & Larcker, 1981). The squared correlation between any two latent variables should be less than the AVE of each variable. This ensures that the latent variables are sufficiently distinct from one another. The Fornell-Larcker criterion is used to assess discriminant validity in structural equation modeling. This multi-stage process ensured that the scale was both reliable and valid, accurately capturing the targeted construct. 3. Results 3.1 Item discrimination The item discrimination analysis was conducted by dividing the sample into two groups based on their total scores: the high-score group (total≥76) and the low-score group (total≤60). An independent samples t-test was conducted to evaluate whether the items could effectively differentiate between the high and low-score groups. The results revealed a significant difference between the groups for all items, with a p -value of 0.01, suggesting that the items successfully distinguished individuals with higher and lower groups. These findings support the discriminative power of the items. 3.2 Item-total correlation Prior Item-total correlation, descriptive statistics were examined to assess data normality. The kurtosis values for all items ranged from 0.152 to 0.745, and the skewness values ranged from 0.193 to 0.475. Since both skewness and kurtosis values were within the acceptable range of ±1, the data were considered approximately normally distributed. Then the item-total correlation was assessed using Pearson’s correlation coefficient, revealing values ranging from 0.791 to 0.878 ( p < 0.001). These results indicate that items demonstrated strong correlations with the total score, suggesting that they are effective indicators of the overall construct. Overall, the item-total correlations support the internal consistency and homogeneity of the scale. 3.3 Cronbach's α if item deleted The overall Cronbach's α for the scale was 0.979, indicating excellent internal consistency. Furthermore, the analysis of Cronbach's α if item deleted revealed that removing any individual item did not significantly improve the overall reliability of the scale. These findings suggest that all items contribute meaningfully to the scale's reliability, and there is no indication that any items should be removed to enhance the internal consistency of the measurement. Therefore, the scale is considered to have strong reliability across all items. 3.4 Exploratory factor analysis To explore the feasibility of conducting Exploratory Factor Analysis (EFA), the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett's Test of Sphericity were first conducted. The KMO value was 0.964, indicating that the data were suitable for factor analysis. Additionally, Bartlett's Test yielded a chi-square ( χ2 ) value of 12,218.747 ( df = 190), with a p -value less than 0.001. This significant result suggests that the correlations among the items were sufficiently strong to justify performing factor analysis. Therefore, the data met the necessary assumptions for conducting EFA. To determine the appropriate number of factors to retain, Parallel Analysis was conducted. Given that the underlying constructs were assumed to be interrelated, an oblique rotation method (Promax) was applied to facilitate interpretability of the factor structure. Parallel analysis compares the eigenvalues from the actual data with those from randomly generated data to determine if the factors retained are meaningful. If the eigenvalues from the actual data exceed those from the random data, the factors are considered significant. Subsequently, based on the following criteria, items were evaluated for deletion: (1) factor loadings below 0.40; (2) communalities below 0.30; (3) items loading on two or more factors with loadings above 0.30; (4) factors with two or fewer items. Parallel analysis conducted in JASP indicated the retention of 4 factors. The results aligned with the pre-specified four-factor model (Regulation, Resonance, Reinforcement, and Reflection) as shown in Table 2, supporting the theoretical structure of the scale and confirming that the factors identified through the analysis were consistent with the expected dimensions, reinforcing the alignment between the qualitative insights and the factor structure derived from the quantitative analysis. Table 2. Exploratory factor analysis results Items Factor1 Factor2 Factor3 Factor4 Uniqueness Communalities Regulation1 0.828 0.257 0.743 Regulation2 0.881 0.203 0.797 Regulation3 0.770 0.207 0.793 Regulation4 0.798 0.207 0.793 Regulation5 0.775 0.170 0.830 Resonance1 0.528 0.248 0.752 Resonance2 0.630 0.233 0.767 Resonance3 0.651 0.199 0.801 Resonance4 0.839 0.148 0.852 Resonance5 0.551 0.233 0.767 Resonance6 0.796 0.172 0.828 Reinforcement1 0.671 0.185 0.815 Reinforcement2 0.769 0.221 0.779 Reinforcement3 0.709 0.214 0.786 Reinforcement4 0.748 0.196 0.804 Reinforcement5 0.807 0.165 0.835 Reflection1 0.731 0.235 0.765 Reflection2 0.846 0.187 0.813 Reflection3 0.610 0.299 0.701 Reflection4 0.830 0.176 0.824 Note: Applied Rotation Method is promax. F actor loadings below 0.4 were not displayed. The Cronbach's alpha coefficients for the four categories in the 4R Model of Generative AI-Based Emotional Support were as follows: Regulation (0.949), Reinforcement (0.953), Resonance (0.957), and Reflection (0.930). As shown in Table 3, the explained variances for each category were 21.2% for Regulation, 21.0% for Reinforcement, 20.7% for Resonance, and 16.3% for Reflection. The cumulative explained variance was 79.2%, indicating that the four categories together account for a substantial portion of the total variance. These values demonstrate high internal consistency and the meaningful contribution of each dimension to the overall scale. Table 3. Factor Characteristics Unrotated solution Rotated solution Eigenvalues SumSq. Loadings Proportion var. Cumulative SumSq. Loadings Proportion var. Cumulative Regulation 14.373 14.166 0.708 0.708 4.245 0.212 0.212 Reinforcement 1.049 0.840 0.042 0.750 4.209 0.210 0.423 Resonance 0.684 0.477 0.024 0.774 4.135 0.207 0.629 Reflection 0.560 0.361 0.018 0.792 3.255 0.163 0.792 3.5 Confirmatory factor analysis We applied SmartPLS (CB-SEM module) to conduct Confirmatory Factor Analysis (CFA) on the second set of data as shown in Figure 1. CFA was a statistical method used to verify if the model proposed by researchers aligned with the actual data. Because the χ2 test statistic was reported, but it was not used as the primary indicator for evaluating model fit. This is because the χ2 statistic is known to be sensitive to sample size, and relying on it alone could lead to inappropriate conclusions. Instead, the focus was on other more robust fit indices, such as RMSEA, SRMR, CFI, and TLI, to comprehensively evaluate the fit between the hypothesized model and the observed data. The thresholds for these indicators typically included CFI and TLI greater than 0.90, and SRMR less than 0.08. RMSEA faces considerable difficulties when assessing simpler models with few df , notably apparent in basic CFAs where df are often limited. In such instances, RMSEA might erroneously indicate a subpar fit, even when the model accurately represents the data (Kenny et al., 2015). Consequently, we adopt a less strict threshold (0.090) to mitigate this issue. 3.6 Model comparison Considering that the partial correlations between the factors were very high, we conducted a model comparison between the four-factor model and the one-factor model. The comparison primarily focused on the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC), where lower values indicate better model fit. In addition, we compared other model fit indices between the four-factor and one-factor models, such as [CFI, TLI, RMSEA, SRMR]. The results in Table 4 showed that the four-factor model outperformed the one-factor model across all fit indices, indicating that the four-factor structure provides a better representation of the data, thereby supporting our research hypothesis. Table 4. Model fit statistics Model χ2 df p χ2/df RMSEA[LOW 90% CI- HIGH 90% CI] SRMR TLI CFI AIC BIC 4-factor model 693.116 164 <0.001 4.226 0.080 [0.074--0.087] 0.022 0.948 0.956 785.116 978.988 1-factor model 1790.283 170 <0.001 10.531 0.138 [0.132-0.144] 0.046 0.848 0.864 1870.283 2038.867 3.7 Convergent validity The assessment of convergent validity was conducted through several key indicators. First, the standardized factor loadings ranged from 0.823 to 0.915, indicating a strong correlation between the measurement items and their corresponding latent variables, which meets established measurement standards. Second, the CR values fell between 0.937 and 0.956, surpassing the conventional threshold of 0.7, thus demonstrating high internal consistency. Additionally, the AVE values ranged from 0.765 to 0.788, exceeding the acceptable level of 0.5, which suggests that a significant proportion of variance is accounted for by the latent variables. Overall, these results indicate that the scale possesses good convergent validity, making it suitable for further research and analysis. The CR values for all factors fell between 0.867 and 0.944, well above the 0.70 recommended standard, suggesting the scale dimensions have strong internal consistency. The AVE values for the factors ranged from 0.621 to 0.771, all meeting the 0.50 threshold for convergent validity. This implies the scale is able to account for a significant portion of the variance in its constituent items. In summary, the comprehensive analysis of factor loadings, CR, and AVE provides evidence of the strong convergent validity of the scale. 3.8 Discriminant validity To establish the discriminant validity of the scale, this study utilized the Fornell-Larcker criterion. The Fornell-Larcker criterion stipulates that the square root of the AVE for each construct should be greater than its correlations with other constructs. In the current analysis as displayed in Table 5, we observe that the square root of the AVE for Reflection (0.888) is lower than its correlation with Reinforcement (0.920), and that square root of the AVE for Resonance (0.884) is lower than its correlation with Reinforcement (0.918). This suggests that Reflection and Reinforcement, as well as Resonance and Reinforcement, are not sufficiently distinct, which violates the Fornell-Larcker criterion for discriminant validity. Table 5. Fornell-Larcker results Reflection Regulation Reinforcement Resonance Reflection 0.888 Regulation 0.824 0.874 Reinforcement 0.920 0.843 0.882 Resonance 0.883 0.853 0.918 0.884 Note: The diagonal values represent the square root of the AVE. 4. Discussion The primary aim of this study was to develop and validate a scale designed to measure the emotional support provided by generative artificial intelligence (AI). This scale, which consists of four key dimensions—Regulation, Resonance, Reinforcement, and Reflection—was created to capture the multifaceted nature of AI-driven emotional support. The development process involved a rigorous validation process, including item analysis, exploratory factor analysis (EFA), and confirmatory factor analysis (CFA). 4.1 Discussion on the rationale of four factors The results from the EFA indicated that the four-factor structure (Regulation, Resonance, Reinforcement, and Reflection) explained a significant portion of the variance in users' perceptions of AI emotional support, with each dimension contributing to the overall model. The four dimensions—Regulation, Resonance, Reinforcement, and Reflection—are grounded in the capabilities of AI to support emotional well-being. Regulation is well represented by items that focus on AI’s ability to help users manage negative emotions, such as through relaxation techniques or providing strategies to cope with stress. These items align with the dimension’s focus on emotional control. Resonance is effectively captured by items that highlight AI's capacity to understand, respond to, and validate users' emotions. This emotional connection is key to helping users feel heard and supported, which is central to this dimension. Reinforcement is reflected in items that emphasize AI's role in providing positive feedback, boosting confidence, and motivating users to continue engaging in positive behaviors. This aligns with the definition of reinforcement as encouraging continued progress and self-belief. Finally, Reflection is represented by items that prompt users to engage in self-reflection about their emotions and behaviors. AI aids users in gaining deeper insights into their emotional responses and behaviors, fostering personal growth, which fits the essence of the Reflection dimension. This alignment between the items and the dimensions they represent supports the accuracy and appropriateness of the chosen labels for each dimension. 4.2 Discussion on the overlap between factors CFA further supported the findings from EFA, demonstrating a good model fit with the data. Additionally, the convergent validity of the scale was established, as each factor showed high internal consistency. Although the four dimensions showed strong internal consistency and good model fit, some concerns arose regarding the discriminant validity of the scale, particularly between the Reflection and Reinforcement dimensions, as well as between Resonance and Reinforcement. The correlations between these dimensions were higher than expected, indicating that they may not be sufficiently distinct from each other. Four possible explanations for the inadequate discriminant validity are: overlapping contextual assumptions, interwoven support types, a sample that lacks diversity, and overlap of multiple response strategies. First, although the items in different dimensions belong to different types of support, they are often designed based on similar contextual assumptions. This means that most items assume that generative AI can recognize and respond to users' emotional states. The similarity in these assumptions may lead to overlap between the items in different dimensions during actual use, making it difficult for respondents to distinguish between these types of support, thereby affecting the discriminant validity between dimensions. Second, these types of support are often not independent. For example, users may experience both emotional regulation and emotional reinforcement during their interaction with AI. In the process of emotional regulation, AI may further enhance the user's emotional state through positive feedback. This interweaving and interaction between different support types may blur the boundaries between dimensions, thus affecting their discriminant validity. Thirdly, the homogeneity of the sample may also impact discriminant validity. If the characteristics of the sample are too concentrated, it could affect the diversity of responses when answering items related to different dimensions. Such homogeneity in the sample may lead to more consistent responses to certain dimensions, reducing the ability to distinguish between them. Fourth, generative AI tools often employ multiple approaches simultaneously when generating responses. These tools combine various algorithms and strategies, which can result in overlapping or intertwined emotional expressions across different dimensions. This multi-faceted response mechanism can blur the boundaries between dimensions, making it difficult to clearly distinguish between them. Therefore, while these dimensions are conceptually distinct, their practical application in generative AI responses may lead to interactions between them that reduce their differentiation. 5. Limitations and suggestions Although this study provides valuable insights, there are several limitations that should be considered. First, although the sample size is sufficient, all the samples are from the same school. Despite the inclusion of students from different academic backgrounds, there may still be issues of homogeneity, which could affect the generalizability of the research findings. Therefore, future studies could consider collecting data from multiple schools or different regions to increase sample diversity and improve the external validity of the findings. Second, this study adopts a cross-sectional design, which means that repeated measurements from the same participants were not collected, and test-retest reliability was not assessed. Therefore, future research could employ a longitudinal design to track changes in participants over time, in order to verify the stability and reliability of the results. Third, this study relied solely on Classical Test Theory for measurement. Future research could incorporate other methodologies, such as network analysis or Item Response Theory (IRT), to provide a more comprehensive validation of the findings. These additional methods could help improve the robustness and depth of the measurement process, offering further insights into the reliability and validity of the results. Finally, this research is specifically focused on Artificial Intelligence Generated Content technology in a general context, without targeting any particular product, e.g., Doubao or ChatGPT. Although the findings shed light on the potential emotional support capabilities of AIGC, they may not be fully applicable to specific applications. Therefore, future research can validate the generalizability of the findings by testing AIGC in different products, examining its effectiveness and emotional support capabilities across various real-world contexts. 6. Implications Given the rising number of mental health issues and the increasing demand for counseling services, the potential of AI in providing emotional support has become more apparent. As more people begin to rely on AI technology for emotional support, it has become essential to measure individuals' perceptions of the emotional support provided by generative AI. The findings of this study have several important implications for both research and practice. First, in the field of psychological counseling, the study highlights how generative artificial intelligence can play a supportive role in emotional regulation, offering an innovative tool to enhance mental health interventions. By integrating AI-based emotional support into counseling sessions, mental health professionals can provide more personalized and responsive care. Second, this study emphasizes how developers can further enhance these features to make emotional support more targeted and effective. By incorporating more advanced emotional intelligence into AI systems, developers can ensure that the AI tools are better at recognizing, resonating with, and responding to users' emotional states. This allows for more personalized and empathetic interactions, improving the user experience and satisfaction. Third, from a practical perspective, by leveraging generative AI to provide immediate emotional responses and guidance, these services can be scaled up, reducing the need for extensive human intervention. This approach makes emotional support more affordable and accessible to a wider range of individuals. In addition, the integration of AI can help protect user privacy by offering emotional support without the need for face-to-face interactions. This could reduce the stigma often associated with seeking psychological help, allowing individuals to access support without the fear of being labeled or judged. By providing anonymous assistance, AI helps maintain privacy while delivering much-needed emotional guidance. 7. Conclusion In conclusion, this study highlights the significant potential of AIGC technology in providing emotional support across various contexts. By exploring its capabilities in emotional regulation, resonance, reinforcement, and reflection, the research offers valuable insights into how AIGC can be used to address the emotional needs of individuals. Despite the limitations, this research paves the way for a deeper exploration of AIGC's integration into emotionally intelligent systems, offering new possibilities for innovation in AI-driven solutions. Declarations Ethics approval and consent to participate This study was approved by the Macau Millennium College Research Ethics Committee (Approval No.: MMCIRB-2024-001). A digital consent form was embedded in the questionnaire, and participants could only proceed after clicking the "Agree to Participate" button. In addition, the participants were fully informed about the study's objectives, procedures, risks, and benefits prior to giving their consent. By clicking the "Agree to Participate" button, they confirmed their understanding and voluntary agreement to take part in the research. The process ensured that all participants provided informed consent electronically before engaging with the survey. All methods were performed in accordance with the Declaration of Helsinki. Consent for publication The embedded digital consent form specified the purpose of the study, and all research data are solely used for academic purposes. All participants were informed that the results of the study may be published in academic journals or presented at conferences. Competing interests The authors declare no competing interests. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Author Contribution YYC: Responsible for research design, data collection, and analysis.YXX and LXY: Responsible for project supervision and final manuscript preparation.WQ: Responsible for experimental implementation and data management.WXF: Responsible for manuscript editing.TZH: Responsible for literature review, data analysis, results discussion, and initial draft writing. Acknowledgement Not applicable. Data Availability The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. References Agnafors, S., Barmark, M. & Sydsjö, G. Mental health and academic performance: a study on selection and causation effects from childhood to early adulthood. Soc. Psychiatry Psychiatr. Epidemiol. 56 , 857–866 (2021). Arifudin, M., Robbaniyah, I. & Ramadhan, I. Confiding through Artificial Intelligence: Can ChatGPT Provide Emotional Support? Proceeding of the International Conference on Social Sciences and Humanities Innovation , 1 (1), 69–78. (2024). Ault, S., Breitenstein, S. M., Tucker, S., Havercamp, S. M. & Ford, J. L. Caregivers of children with autism spectrum disorder in rural areas: A literature review of mental health and social support. J. Pediatr. Nurs. 61 , 229–239 (2021). Barratt, J. M. & Duran, F. Does psychological capital and social support impact engagement and burnout in online distance learning students? Internet High. Educ. 51 , 100821 (2021). Bearman, M. & Ajjawi, R. Learning to work with the black box: Pedagogy for a world with artificial intelligence. Br. J. Edu. Technol. 54 (5), 1160–1173 (2023). Bedaso, A., Adams, J., Peng, W. & Sibbritt, D. The relationship between social support and mental health problems during pregnancy: a systematic review and meta-analysis. Reproductive Health . 18 , 1–23 (2021). Buck, D. et al. M., Impact of emotional support for mental health student nurses and retention on the mental health nursing programme. [Vrije Universiteit Amsterdam]. (2024). https://doi.org/10.5463/thesis.859 Cheng Lin, K., Huang, T. C., Hung, J. C., Yen, N. Y., Chen, J. & S Facial emotion recognition towards affective computing-based learning. Libr. Hi Tech. 31 (2), 294–307. https://doi.org/10.1108/07378831311329068 (2013). Chu, T., Liu, X., Takayanagi, S., Matsushita, T. & Kishimoto, H. Association between mental health and academic performance among university undergraduates: The interacting role of lifestyle behaviors. International Journal of Methods in Psychiatric Research , 32 (1), e1938. (2023). Díaz, Y., Hessel, P., Avendano, M. & Evans-Lacko, S. Multidimensional poverty and adolescent mental health: Unpacking the relationship. Soc. Sci. Med. 311 , 115324 (2022). Duncan, M. J., Patte, K. A. & Leatherdale, S. T. Mental health associations with academic performance and education behaviors in Canadian secondary school students. Can. J. School Psychol. 36 (4), 335–357 (2021). Etori, N. A. & Gini, M. WisCompanion: Integrating the Socratic Method with ChatGPT-Based AI for Enhanced Explainability in Emotional Support for Older Adults. International Conference on Human-Computer Interaction , 179–198. (2024). Fornell, C. & Larcker, D. F. Evaluating structural equation models with unobservable variables and measurement error. J. Mark. Res. 18 (1), 39–50. https://doi.org/10.2307/3151312 (1981). Gaspar, T., Cerqueira, A., Guedes, F. B. & de Matos, M. G. Parental Emotional Support, Family Functioning and Children’s Quality of Life. Psychol. Stud. 67 (2), 189–199 (2022). Gibson, B., Schneider, J., Talamonti, D. & Forshaw, M. The impact of inequality on mental health outcomes during the COVID-19 pandemic: A systematic review. Can. Psychol. 62 (1), 101 (2021). Gilbert, M. H., Dextras-Gauthier, J., St-Hilaire, F. & Lerouge, L. Understanding managers’ mental health: the cornerstone for better organizational performance and workers’ health. In Frontiers in Psychology (Vol. 15, 1468825). Frontiers Media SA. (2024). Goel, N. J., Thomas, B., Boutté, R. L., Kaur, B. & Mazzeo, S. E. What will people say? Mental health stigmatization as a barrier to eating disorder treatment-seeking for South Asian American women. Asian Am. J. Psychol. 14 (1), 96 (2023). Guo, K. et al. Assessing social support impact on depression, anxiety, and stress among undergraduate students in Shaanxi province during the COVID-19 pandemic of China. PloS One . 16 (7), e0253891 (2021). Hair, J., Black, W., Babin, B. & Anderson, R. Multivariate data analysis (7th Editio). Prentice-Hall, Inc. (2010). Hassija, V. et al. Interpreting black-box models: a review on explainable artificial intelligence. Cogn. Comput. 16 (1), 45–74 (2024). Herbener, A. B. & Damholdt, M. F. Are lonely youngsters turning to chatbots for companionship? The relationship between chatbot usage and social connectedness in Danish high-school students. Int. J. Hum. Comput. Stud. 196 , 103409 (2025). Hinkson, K. D. Jr et al. An examination of the mental health and academic performance of student veterans. J. Am. Coll. Health . 70 (8), 2334–2341 (2022). Holliman, A. J. et al. Adaptability and social support: Examining links with psychological wellbeing among UK students and non-students. Front. Psychol. 12 , 636520 (2021). House, J. S. Work stress and social support (Addison-Wesley Series on Occupational Stress, 1983). Hu, L. T. & Bentler, P. M. Cutoff Criteria for Fit Indexes in Covariance Structure Analysis: Conventional Criteria Versus New Alternatives. Struct. Equation Modeling: Multidisciplinary J. 6 (1), 1–55. https://doi.org/10.1080/10705519909540118 (1999). Hu, X. et al. Understanding the impact of emotional support on mental health resilience of the community in the social media in Covid-19 pandemic. J. Affect. Disord. 308 , 360–368 (2022). Huang, L. & Zhang, T. Perceived social support, psychological capital, and subjective well-being among college students in the context of online learning during the COVID-19 pandemic. Asia-Pacific Educ. Researcher . 31 (5), 563–574 (2022). Huang, Y. et al. The impacts of coping style and perceived social support on the mental health of undergraduate students during the early phases of the COVID-19 pandemic in China: a multicenter survey. BMC Psychiatry . 21 , 1–12 (2021). Jauch, M., Occhipinti, S. & O’Donovan, A. The stigmatization of mental illness by mental health professionals: Scoping review and bibliometric analysis. PLoS One , 18 (1), e0280739. (2023). Kaiser, H. F. An index of factorial simplicity. Psychometrika 39 (1), 31–36. https://doi.org/10.1007/BF02291575 (1974). Karadaş, A. & Duran, S. The effect of social support on work stress in health workers during the pandemic: The mediation role of resilience. J. Community Psychol. 50 (3), 1640–1649 (2022). Kenny, D. A., Kaniskan, B. & McCoach, D. B. The Performance of RMSEA in Models With Small Degrees of Freedom. Sociol. Methods Res. 44 (3), 486–507. https://doi.org/10.1177/0049124114543236 (2015). Kikas, E. & Tang, X. Child-perceived teacher emotional support, its relations with teaching practices, and task persistence. Eur. J. Psychol. Educ. 34 , 359–374 (2019). Li, W. et al. Socioeconomic inequality in child mental health during the COVID-19 pandemic: First evidence from China. J. Affect. Disord. 287 , 8–14 (2021). Li, X. Y. The relationship between quality of sports friendships and mental health in Chinese junior high school students: the bidirectional chain mediating effects of sport motivation and exercise adherence. BMC Public. Health . 25 (1), 52. https://doi.org/10.1186/s12889-025-21287-5 (2025). Loiacono, E. T., Watson, R. T. & Goodhue, D. L. WebQual: A measure of website quality. Mark. Theory Appl. 13 (3), 432–438 (2002). Mai, Y., Wu, Y. J. & Huang, Y. What type of social support is important for student resilience during COVID-19? A latent profile analysis. Front. Psychol. 12 , 646145 (2021). McDonald, R. P. & Ho, M. H. R. Principles and practice in reporting structural equation analyses. Psychol. Methods . 7 (1), 64–82. https://doi.org/10.1037/1082-989X.7.1.64 (2002). Mendoza-Pinto, R. Artificial Intelligence in the Fight Against Bullying: Integration of ChatGPT in an Emotional Support Chatbot. Proceedings Http://Ceur-Ws. Org ISSN , 1613 , 73. (2023). Moore, P. Do We Understand the Relationship between Affective Computing, Emotion and Context-Awareness? Machines (Basel) . 5 (3), 16. https://doi.org/10.3390/machines5030016 (2017). Naher, J. Can ChatGPT provide a better support: a comparative analysis of ChatGPT and dataset responses in mental health dialogues. Current Psychology , 1–9. (2024). O’Connor, R. C. et al. Mental health and well-being during the COVID-19 pandemic: longitudinal analyses of adults in the UK COVID-19 Mental Health & Wellbeing study. Br. J. Psychiatry . 218 (6), 326–333 (2021). Pei, G. et al. Affective Computing: Recent Advances, Challenges, and Future Trends. Intelligent Computing , 3 . (2024). https://doi.org/10.34133/icomputing.0076 Permatasari, N., Ashari, F. R. & Ismail, N. Contribution of perceived social support (peer, family, and teacher) to academic resilience during COVID-19. Gold. Ratio Social Sci. Educ. 1 (1), 1–12 (2021). Picard, R. W. Affective computing (MIT Press, 2000). Polanin, J. R. et al. A meta-analysis of longitudinal partial correlations between school violence and mental health, school performance, and criminal or delinquent acts. Psychol. Bull. 147 (2), 115 (2021). Rashid Soron, T. et al. H., C Domestic violence and mental health during the COVID-19 pandemic in Bangladesh. JMIR Formative Res. , 5 (9), e24624. (2021). Rider, E. A., Ansari, E., Varrin, P. H. & Sparrow, J. Mental health and wellbeing of children and adolescents during the covid-19 pandemic. Bmj , 374 . (2021). Rohan, R., Faruk, L. I. D., Puapholthep, K. & Pal, D. Unlocking the Black Box: Exploring the use of Generative AI (ChatGPT) in Information Systems Research. Proceedings of the 13th International Conference on Advances in Information Technology , 1–9. (2023). Romm, K. F., Metzger, A. & Turiano, N. A. Parental emotional support and health problems: The role of social support and social strain. J. Adult Dev. 28 (4), 319–331 (2021). Schenke, K., Ruzek, E., Lam, A. C., Karabenick, S. A. & Eccles, J. S. To the means and beyond: Understanding variation in students’ perceptions of teacher emotional support. Learn. Instruction . 55 , 13–21 (2018). Shin, D. et al. Exploring the effects of AI-assisted emotional support processes in online mental health community. CHI Conference on Human Factors in Computing Systems Extended Abstracts , 1–7. (2022). Shuo, Z., Xuyang, D., Xin, Z., Xuebin, C. & Jie, H. The relationship between postgraduates’ emotional intelligence and well-being: the chain mediating effect of social support and psychological resilience. Front. Psychol. 13 , 865025 (2022). Steiger, S. et al. Personality, self-esteem, familiarity, and mental health stigmatization: a cross-sectional vignette-based study. Sci. Rep. 12 (1), 10347 (2022). Thomson, R. M., Kopasker, D., Leyland, A., Pearce, A. & Katikireddi, S. V. Effects of poverty on mental health in the UK working-age population: causal analyses of the UK Household Longitudinal Study. Int. J. Epidemiol. 52 (2), 512–522 (2023). Tibber, M. S., Walji, F., Kirkbride, J. B. & Huddy, V. The association between income inequality and adult mental health at the subnational level—a systematic review. Social Psychiatry Psychiatric Epidemiology , 1–24. (2022). Wang, Y. et al. A systematic review on affective computing: emotion models, databases, and recent advances. Inform. Fusion . 83–84 , 19–52. https://doi.org/10.1016/j.inffus.2022.03.009 (2022). WHO. Mental health . World Health Organization. (2022). https://www.who.int/news-room/fact-sheets/detail/mental-health-strengthening-our-response Wolfinbarger, M. & Gilly, M. C. eTailQ: dimensionalizing, measuring and predicting etail quality. J. Retail. 79 (3), 183–198 (2003). Yeung, R. & Leadbeater, B. Adults make a difference: the protective effects of parent and teacher emotional support on emotional and behavioral problems of peer-victimized adolescents. J. Community Psychol. 38 (1), 80–98 (2010). Zaneva, M., Guzman-Holst, C., Reeves, A. & Bowes, L. The impact of monetary poverty alleviation programs on children’s and adolescents’ mental health: A systematic review and meta-analysis across low-, middle-, and high-income countries. J. Adolesc. Health . 71 (2), 147–156 (2022). Zeng, Z. Emotional Support and Mental Health: Investigating Interpersonal Emotion Regulation in Chinese and Dutch Contexts . (2024). Zhou, Z. & Cheng, Q. Relationship between online social support and adolescents’ mental health: A systematic review and meta-analysis. J. Adolesc. 94 (3), 281–292 (2022). Additional Declarations No competing interests reported. Supplementary Files AppendixA.docx 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-6956402","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":489196640,"identity":"de6a4575-320f-4dce-a5e6-52f14b81697b","order_by":0,"name":"Yanchao Yang","email":"","orcid":"","institution":"Macau Millennium College","correspondingAuthor":false,"prefix":"","firstName":"Yanchao","middleName":"","lastName":"Yang","suffix":""},{"id":489196641,"identity":"0424550a-eb5d-4e96-9269-dbb8d4dd72fb","order_by":1,"name":"Xinxin Yang","email":"","orcid":"","institution":"Macau Millennium College","correspondingAuthor":false,"prefix":"","firstName":"Xinxin","middleName":"","lastName":"Yang","suffix":""},{"id":489196642,"identity":"dbf2060a-4aa5-46d4-97c1-ff9f91e5bb78","order_by":2,"name":"Zehang Tan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAs0lEQVRIiWNgGAWjYDACCRBRgcQmQgszkDhDshbGNlK0yM/uP/i5cN62aP4G5oO3eRi2yRHUwjjnMLP0zG23c2ccYEu25mG4bUxQC7NEMoM0L1DLBgYeM2mglsQGQlrYJJKZf/POAWnh/0acFh6JZDZp3gawLWzEaZGQSDaz5jkG9MthNmPLOQZE+EV+RuLj2zw1t3P725sf3nhTcZtwiCEAKEoZDEjQMApGwSgYBaMANwAAK340qM/MExYAAAAASUVORK5CYII=","orcid":"","institution":"Macau Millennium College","correspondingAuthor":true,"prefix":"","firstName":"Zehang","middleName":"","lastName":"Tan","suffix":""},{"id":489196643,"identity":"d42b5c5d-74fc-4fdb-8828-d556f28a00f2","order_by":3,"name":"Qiu Wei","email":"","orcid":"","institution":"Macau Millennium College","correspondingAuthor":false,"prefix":"","firstName":"Qiu","middleName":"","lastName":"Wei","suffix":""},{"id":489196644,"identity":"b60a435f-2896-4c43-91d7-660ebc164ad7","order_by":4,"name":"Xiaofeng Wang","email":"","orcid":"","institution":"University of Macau","correspondingAuthor":false,"prefix":"","firstName":"Xiaofeng","middleName":"","lastName":"Wang","suffix":""},{"id":489196645,"identity":"fda7ce8b-6bab-4b46-91d0-144c1a694002","order_by":5,"name":"Xinyu Liu","email":"","orcid":"","institution":"Macau Millennium College","correspondingAuthor":false,"prefix":"","firstName":"Xinyu","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2025-06-23 11:38:27","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6956402/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6956402/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87576111,"identity":"3869de9e-006f-41ca-a204-e8fa0bad5b08","added_by":"auto","created_at":"2025-07-25 11:39:48","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":179800,"visible":true,"origin":"","legend":"\u003cp\u003e4R Model of Generative AI-Based Emotional Support Scale\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6956402/v1/3fe5b9e043dd69693528a12b.png"},{"id":92159533,"identity":"6152089e-a86a-4eb1-9fe2-638cea7683d7","added_by":"auto","created_at":"2025-09-25 09:47:12","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1078735,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6956402/v1/160f17f9-f663-4a44-ad6b-008452aaa6f4.pdf"},{"id":87577044,"identity":"d543a2fa-d2aa-441a-89fa-fce056c74594","added_by":"auto","created_at":"2025-07-25 11:47:48","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":18794,"visible":true,"origin":"","legend":"","description":"","filename":"AppendixA.docx","url":"https://assets-eu.researchsquare.com/files/rs-6956402/v1/d7d4cda0ea38f0e217f05a05.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Conceptualizing and Validating the 4R Model of Generative AI-Based Emotional Support Scale: Regulation, Resonance, Reinforcement, and Reflection","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eMental health, referring to a state of emotional and psychological well-being that allows individuals to manage life's challenges, recognize their strengths, perform effectively in learning and work environments, and positively contribute to their communities (WHO, 2022), has consistently been a research focus in academic fields (X.-Y. Li, 2025). Many studies have delved into the antecedents of mental health issues, investigating a range of factors such as social and economic conditions, including poverty (D\u0026iacute;az et al., 2022; Thomson et al., 2023; Zaneva et al., 2022), violence (Polanin et al., 2021; Rashid Soron et al., 2021), and inequality (Gibson et al., 2021; W. Li et al., 2021; Tibber et al., 2022), and adverse experiences during critical periods. Concurrently, a substantial body of research has explored the consequences of mental health on various domains of life, including academic performance (Agnafors et al., 2021; Chu et al., 2023; Duncan et al., 2021; Hinkson Jr et al., 2022), wellbeing (O\u0026rsquo;Connor et al., 2021; Rider et al., 2021) and organizational performance (Gilbert et al., 2024), demonstrating its profound influence on individuals' capacity to function and thrive in both personal and professional contexts.\u003c/p\u003e\u003cp\u003eOne important area of research in mental health is exploring how to improve mental health through social support (Ault et al., 2021; Bedaso et al., 2021; Y. Huang et al., 2021; Zhou \u0026amp; Cheng, 2022). Social support is a multifaceted concept that has garnered significant attention across various fields of research, particularly in psychology, education, and healthcare. It generally refers to the assistance and resources provided by others\u0026mdash;whether individuals, groups, or organizations\u0026mdash;that help individuals manage stress (Guo et al., 2021; Karadaş \u0026amp; Duran, 2022), foster resilience (Mai et al., 2021; Permatasari et al., 2021), and enhance their well-being (Holliman et al., 2021; L. Huang \u0026amp; Zhang, 2022; Shuo et al., 2022) and learning engagement (Barratt \u0026amp; Duran, 2021). Perceived social support can be divided into four key dimensions: emotional support, informational support, instrumental support, and appraisal support (House, 1983). Among these, emotional support stands out as a crucial factor in mental health, as it involves providing empathy, care, and understanding during times of distress. Emotional support from family, friends, and even colleagues can significantly reduce the impact of stress, increase feelings of security, and enhance emotional resilience. The presence of emotional support, whether in family settings or organizational contexts, has been shown to buffer individuals against mental health challenges, facilitating better overall well-being and functioning (Buck et al., 2024; X. Hu et al., 2022; Zeng, 2024).\u003c/p\u003e\u003cp\u003eCurrently, emotional support is predominantly provided by individuals such as teachers (Kikas \u0026amp; Tang, 2019; Schenke et al., 2018; Yeung \u0026amp; Leadbeater, 2010) and parents (Gaspar et al., 2022; Romm et al., 2021). However, individuals experiencing mental health challenges often face the fear of being stigmatized or labeled (Goel et al., 2023; Jauch et al., 2023; Steiger et al., 2022), which can act as a significant barrier to seeking traditional forms of support. This stigma can lead to reluctance in disclosing one's struggles, thus exacerbating feelings of isolation and distress. Consequently, the need for confidential and non-judgmental emotional support has become increasingly critical for individuals facing mental health issues. Human-chatbot interactions have indeed evolved significantly over time. The emergence of AI-generated content (AIGC) technologies presents new opportunities to address this gap, offering individuals the ability to access emotional support in a private and anonymous manner (Arifudin et al., 2024; Mendoza-Pinto, 2023), thanks to affective computing (Picard, 2000). Affective computing can perform (a) emotion recognition (Cheng Lin et al., 2013), which involves text sentiment analysis, speech emotion recognition, facial expression recognition, and physiological signal processing; (b) emotion understanding (Moore, 2017; Wang et al., 2022), which focuses on developing context-aware systems; and (c) emotion generation (Pei et al., 2024), which includes emotional speech synthesis, therapeutic music generation, emotional visual content creation, and dialogue generation systems.\u003c/p\u003e\u003cp\u003eRecent empirical research has begun to investigate the role of AIGC in delivering personalized, confidential emotional support, with promising findings suggesting that it can facilitate improved mental well-being and mitigate the stigma associated with seeking help (Etori \u0026amp; Gini, 2024; Herbener \u0026amp; Damholdt, 2025; Naher, 2024; Shin et al., 2022). These studies underscore the potential of AIGC to serve as a valuable resource for enhancing emotional resilience and promoting mental health, particularly in environments where traditional support systems may not be readily accessible or where privacy is a key concern.\u003c/p\u003e\u003cp\u003eGenerative Artificial Intelligence has been widely applied in emotional support scenarios, offering anonymous and personalized emotional care. However, AIGC itself is often regarded as a \"black-box\" system (Bearman \u0026amp; Ajjawi, 2023; Hassija et al., 2024; Rohan et al., 2023), meaning that its inner workings and decision-making processes are not easily understood by external observers. Therefore, when further improving these models to effectively provide emotional support, it becomes crucial to accurately differentiate between various types of emotional support. Thus, this paper primarily constructs and validates a multidimensional emotional support model to optimize the application of AIGC in emotional support. Through this model, we can identify and distinguish the different dimensions of emotional support, such as emotional resonance, emotional regulation, non-judgmental listening, and emotional enhancement. This provides clear guidance for AIGC systems, ensuring that their responses can more precisely align with users' emotional needs.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Ethical considerations\u003c/h2\u003e\u003cp\u003ePrior to data collection, ethical approval was obtained from the Research Ethics Committee of the corresponding author's institution (Approval No. MMCIRB-2024-001). The application outlined the study\u0026rsquo;s aims, methods, privacy protections, and participant criteria, and confirmed that no identifiable information (e.g., names or student IDs) would be collected. Data were gathered via convenience sampling at an independent college in northern China. The survey link, which included an electronic informed consent form, was distributed through instructors via WeChat. Participation was voluntary, and students had to click \"Agree to Participate\" to proceed. All data were anonymized and securely stored for research use only. All methods were performed in accordance with the Declaration of Helsinki.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Participants\u003c/h2\u003e\u003cp\u003eData collection took place in December 2024, with each participant requiring approximately 5 minutes to complete the survey. To ensure data quality, the collected data were filtered based on response time and Mahalanobis distance, resulting in 996 valid participants. The sample was then randomly divided into two groups for statistical analysis. As shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the first group, used for exploratory factor analysis (EFA), consisted of 496 participants, including 179 males (36.10%) and 317 females (63.90%), with 103 participants (20.80%) from urban areas and 393 (79.20%) from rural areas. The second group, used for confirmatory factor analysis (CFA), included 500 participants, comprising 174 males (34.80%) and 326 females (65.20%), with 117 participants (23.40%) from urban areas and 383 (76.60%) from rural areas.\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\u003eDemographic information\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\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e\u003cp\u003eEFA\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e\u003cp\u003eCFA\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVariable\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGroup\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003eFaculty\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFEEE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e61\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e12.30%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e14.00%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e127\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e25.60%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e122\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e24.40%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFIL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e93\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e18.80%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e16.40%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFB\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e30\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e6.00%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e30\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e6.00%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFPE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e22\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.40%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e20\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e4.00%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFFTAD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e163\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e32.90%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e176\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e35.20%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eSex\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e179\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e36.10%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e174\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e34.80%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e317\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e63.90%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e326\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e65.20%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eBirthplace\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eUrban\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e103\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e20.80%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e117\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e23.40%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRural\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e393\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e79.20%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e383\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e76.60%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"6\"\u003e\u003cem\u003eNote: FEEE\u0026thinsp;=\u0026thinsp;Faculty of Electrical and Information Engineering, FE\u0026thinsp;=\u0026thinsp;Faculty of Engineering, FIL\u0026thinsp;=\u0026thinsp;Faculty of International Languages, FB\u0026thinsp;=\u0026thinsp;Faculty of Business, FPE\u0026thinsp;=\u0026thinsp;Faculty of Physical Education, FFTAD\u0026thinsp;=\u0026thinsp;Faculty of Film, Television, and Art Design\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Instruments\u003c/h2\u003e\u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\u003ch2\u003e2.3.1 4R Model of Generative AI-Based Emotional Support Scale\u003c/h2\u003e\u003cp\u003eThe 4R Model of Generative AI-Based Emotional Support was developed to explore how generative AI can provide effective emotional support. The model is based on expert interviews, qualitative analysis, and rigorous validation.\u003c/p\u003e\u003cp\u003eFirst, five experts were interviewed, including one IT specialist, the former director of a big data center at a Fortune 500 company in China with expertise in AI, an educational measurement expert with multiple SSCI-indexed publications on scale development, an assistant professor with research interest on healing, and a psychology doctoral student. The protocol was developed through an iterative process involving a review of existing literature on emotional support, generative AI interaction, and human-computer communication. The aim was to elicit expert insights on how generative AI can provide emotional support in realistic user-AI interactions. Sample questions were like \u0026ldquo;When users seek emotional or psychological help, how does Gen-AI typically respond\u0026rdquo;.\u003c/p\u003e\u003cp\u003eSecond, the interview for each one lasted approximately 30 minutes, and the interview protocol focused on the potential of generative AI to support emotional well-being, emphasizing its mechanisms and practical applications.\u003c/p\u003e\u003cp\u003eThird, the interview recordings were first transcribed, ensuring that every response was captured accurately. Once transcriptions were completed, the data was analyzed using MAXQDA software, which allowed for in-depth qualitative analysis. Following grounded theory methodology, the analysis was conducted in three stages. To begin with, during initial coding, the transcripts were reviewed line by line, with open coding applied to identify key themes and concepts related to emotional support. Each relevant statement was assigned a preliminary code. Next, in the axial coding stage, the initial codes were grouped into categories based on similarities and relationships, refining the themes related to how generative AI can provide emotional support. Finally, during selective coding, the most significant categories were identified and related to the central research question, leading to the organization and definition of four core categories: Regulation (5 items), Resonance (6 items), Reinforcement (5 items), and Reflection (4 items). Specifically, Regulation refers to AI's ability to help users manage their emotions by offering tools and strategies to cope with stress or negative feelings, such as relaxation techniques One sample item is \u0026ldquo;Generative artificial intelligence guides me to express my emotions, helping me release them through conversation or other means.\u0026rdquo;. Resonance involves AI creating an emotional connection by understanding and responding to users' emotions, making them feel heard and validated. One sample item is \u0026ldquo;When I confide in generative artificial intelligence, it always maintains an open mindset and does not judge my emotions or experiences.\u0026rdquo; Reinforcement is AI\u0026rsquo;s provision of positive feedback and encouragement, which boosts users' confidence and motivates them to continue positive behaviors. One sample item is \u0026rdquo; Whenever I achieve a small accomplishment, generative artificial intelligence acknowledges my efforts and uses positive words to boost my confidence.\u0026rdquo; Finally, Reflection emphasizes AI's role in prompting users to reflect on their emotions and behaviors, aiding them in gaining deeper insights into their feelings and reactions for personal growth. One sample item is \u0026ldquo;Generative artificial intelligence helps me review my emotions and analyze what events or situations triggered them, so I can understand why I have these feelings.\u0026rdquo;\u003c/p\u003e\u003cp\u003eFourth, this model was further validated through expert evaluations, where eight university counselors with backgrounds in psychology, education, and student mental health counseling assessed the scale items for clarity, representativeness, and relevance. The item-level content validity index (I-CVI) and scale-level content validity index (S-CVI) both exceeded 0.9, confirming the scale's strong validity. The final scale adopted a five-point Likert format, where higher scores indicated stronger agreement with the described dimensions, ensuring alignment with user experiences and interpretability.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Analytical procedure\u003c/h2\u003e\u003cp\u003eThe scale development process began with a comprehensive item analysis to assess the discriminatory power of each scale item. Participants were divided into two groups: the top 27% and the bottom 27%, based on their overall scores, enabling a comparison of how well the items differentiated individuals with high versus low levels of the targeted construct. To analyze these differences, an independent samples t-test was conducted for each item. The t-test allowed for comparison of the mean scores between the two groups, with the 95% confidence interval (CI) being used to assess the precision of the estimated difference in means. Items that demonstrated significant differences between the high and low groups, as indicated by a \u003cem\u003ep\u003c/em\u003e-value less than 0.05 and a non-overlapping 95% CI, were retained for the final scale. Conversely, items that did not exhibit significant differences (non-significant t-test results and overlapping 95% CI) were excluded, ensuring that only those items with strong discriminatory power were incorporated into the scale.\u003c/p\u003e\u003cp\u003eThe next step in the scale development process involved item-total correlation analysis, which served as a basis for evaluating the homogeneity of individual items. In this analysis, the correlation between each item and the total score was computed. Items that exhibited lower correlations with the total score were considered to have lower homogeneity with the overall construct and were flagged for potential removal. A threshold of 0.4 was set for the item-total correlation (Loiacono et al., 2002; Wolfinbarger \u0026amp; Gilly, 2003); items with a correlation below this value were deemed to lack sufficient homogeneity and were deleted from the scale. Conversely, items with higher correlations were retained, as they demonstrated strong alignment with the overall construct being measured.\u003c/p\u003e\u003cp\u003eThe analysis of \"Cronbach's α if item deleted\" was also conducted to further assess each item's contribution to the internal consistency of the scale. If deleting an item significantly increased the Cronbach's α, it indicated the item contributed little to consistency and could be removed. Conversely, if the Cronbach's α decreased upon removal, the item was essential for the scale's internal consistency and should be retained.\u003c/p\u003e\u003cp\u003eThe Kaiser-Meyer-Olkin (KMO) measure and Bartlett\u0026rsquo;s Test of Sphericity were used via JASP to assess whether the data were suitable for factor analysis. A KMO value closer to 1 indicated that the data were suitable for factor analysis, with a value greater than 0.8 generally considered acceptable (Kaiser, 1974). If the KMO value was low, it suggested that the correlations between variables were insufficient for factor analysis. Bartlett\u0026rsquo;s Test of Sphericity checked whether the correlation matrix was an identity matrix. A significant result (\u003cem\u003ep\u003c/em\u003e-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05) indicated that there were sufficient correlations among the variables, making it appropriate to proceed with factor analysis. Following this, exploratory factor analysis (EFA) was performed via JASP to identify the latent factor structure underlying the scale. EFA revealed the key factors that explained the most variance in the data, providing insight into how items clustered into coherent dimensions.\u003c/p\u003e\u003cp\u003eOnce the factor structure was established, confirmatory factor analysis (CFA) was conducted to validate the model with SmartPLS (CB-SEM module). A series of fit indices, including RMSEA, SRMR, CFI, TLI, and \u003cem\u003eχ2\u003c/em\u003e statistics with degrees of freedom (\u003cem\u003edf\u003c/em\u003e), were used to assess model fit. A well-fitting model was characterized by RMSEA and SRMR values below 0.08, and CFI and TLI values above 0.90 (Hu \u0026amp; Bentler, 1999; McDonald \u0026amp; Ho, 2002).\u003c/p\u003e\u003cp\u003eFinally, the scale\u0026rsquo;s construct validity was rigorously evaluated through convergent and discriminant validity analyses.\u003c/p\u003e\u003cp\u003eTo assess convergent validity, three key indicators were used: factor loadings, composite reliability (CR), and average variance extracted (AVE) (Fornell \u0026amp; Larcker, 1981; Hair et al., 2010). First, factor loadings were examined. These loadings represent the correlation between each item and its associated factor. A high factor loading (ideally greater than 0.7) suggests that the item is strongly related to the underlying construct it is meant to measure. Items with lower factor loadings were considered less relevant and were potentially removed to ensure the scale measures the intended concept accurately. Second, CR was calculated. CR evaluates the internal consistency of a construct, which refers to how reliably the items within that construct measure the same underlying idea. A CR value above 0.7 indicates acceptable internal consistency, suggesting that the items are cohesive in their measurement of the construct. Finally, AVE was assessed to determine how much variance in the items is explained by the factor, compared to the error variance. An AVE value of 0.5 or higher indicates that more than half of the variance in the items is explained by the construct, thus confirming that the items are valid indicators of the construct.\u003c/p\u003e\u003cp\u003eDiscriminant validity refers to the extent to which a construct is truly distinct from other constructs in the model, ensuring that it does not overlap significantly with other constructs. One common method for assessing discriminant validity is the Fornell-Larcker criterion (Fornell \u0026amp; Larcker, 1981). The squared correlation between any two latent variables should be less than the AVE of each variable. This ensures that the latent variables are sufficiently distinct from one another. The Fornell-Larcker criterion is used to assess discriminant validity in structural equation modeling.\u003c/p\u003e\u003cp\u003eThis multi-stage process ensured that the scale was both reliable and valid, accurately capturing the targeted construct.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003e\u003cstrong\u003e3.1 Item discrimination\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe item discrimination analysis was conducted by dividing the sample into two groups based on their total scores: the high-score group (total\u0026ge;76) and the low-score group (total\u0026le;60). An independent samples t-test was conducted to evaluate whether the items could effectively differentiate between the high and low-score groups. The results revealed a significant difference between the groups for all items, with a\u003cem\u003e\u0026nbsp;p\u003c/em\u003e-value of 0.01, suggesting that the items successfully distinguished individuals with higher and lower groups. These findings support the discriminative power of the items.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Item-total correlation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePrior Item-total correlation, descriptive statistics were examined to assess data normality. The kurtosis values for all items ranged from 0.152 to 0.745, and the skewness values ranged from 0.193 to 0.475. Since both skewness and kurtosis values were within the acceptable range of \u0026plusmn;1, the data were considered approximately normally distributed. Then the item-total correlation was assessed using Pearson\u0026rsquo;s correlation coefficient, revealing values ranging from 0.791 to 0.878 (\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001). These results indicate that items demonstrated strong correlations with the total score, suggesting that they are effective indicators of the overall construct. Overall, the item-total correlations support the internal consistency and homogeneity of the scale.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Cronbach\u0026apos;s \u0026alpha; if item deleted\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe overall Cronbach\u0026apos;s \u0026alpha; for the scale was 0.979, indicating excellent internal consistency. Furthermore, the analysis of Cronbach\u0026apos;s \u0026alpha; if item deleted revealed that removing any individual item did not significantly improve the overall reliability of the scale. These findings suggest that all items contribute meaningfully to the scale\u0026apos;s reliability, and there is no indication that any items should be removed to enhance the internal consistency of the measurement. Therefore, the scale is considered to have strong reliability across all items.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Exploratory factor analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo explore the feasibility of conducting Exploratory Factor Analysis (EFA), the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett\u0026apos;s Test of Sphericity were first conducted. The KMO value was 0.964, indicating that the data were suitable for factor analysis. Additionally, Bartlett\u0026apos;s Test yielded a chi-square (\u003cem\u003e\u0026chi;2\u003c/em\u003e) value of 12,218.747 (\u003cem\u003edf\u0026nbsp;\u003c/em\u003e= 190), with a \u003cem\u003ep\u003c/em\u003e-value less than 0.001. This significant result suggests that the correlations among the items were sufficiently strong to justify performing factor analysis. Therefore, the data met the necessary assumptions for conducting EFA.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo determine the appropriate number of factors to retain, Parallel Analysis was conducted. Given that the underlying constructs were assumed to be interrelated, an oblique rotation method (Promax) was applied to facilitate interpretability of the factor structure. Parallel analysis compares the eigenvalues from the actual data with those from randomly generated data to determine if the factors retained are meaningful. If the eigenvalues from the actual data exceed those from the random data, the factors are considered significant.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSubsequently, based on the following criteria, items were evaluated for deletion: (1) factor loadings below 0.40; (2) communalities below 0.30; (3) items loading on two or more factors with loadings above 0.30; (4) factors with two or fewer items. Parallel analysis conducted in JASP indicated the retention of 4 factors. The results aligned with the pre-specified four-factor model (Regulation, Resonance, Reinforcement, and Reflection) as shown in Table 2, supporting the theoretical structure of the scale and confirming that the factors identified through the analysis were consistent with the expected dimensions, reinforcing the alignment between the qualitative insights and the factor structure derived from the quantitative analysis.\u003c/p\u003e\n\u003cp\u003eTable 2. Exploratory factor analysis results\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eItems\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003eFactor1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003eFactor2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003eFactor3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\n \u003cp\u003eFactor4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003eUniqueness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003eCommunalities\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eRegulation1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.828\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.257\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.743\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eRegulation2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.881\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.203\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.797\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eRegulation3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.770\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.207\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.793\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eRegulation4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.798\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.207\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.793\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eRegulation5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.775\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.170\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.830\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eResonance1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.528\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.248\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.752\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eResonance2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.630\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.233\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.767\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eResonance3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.651\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.199\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.801\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eResonance4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.839\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.148\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.852\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eResonance5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.551\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.233\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.767\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eResonance6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.796\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.172\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.828\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReinforcement1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.671\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.185\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.815\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReinforcement2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.769\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.221\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.779\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReinforcement3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.709\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.214\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.786\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReinforcement4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.748\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.196\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.804\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReinforcement5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\n \u003cp\u003e0.807\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.165\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.835\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReflection1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\n \u003cp\u003e0.731\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.235\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.765\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReflection2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\n \u003cp\u003e0.846\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.187\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.813\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReflection3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\n \u003cp\u003e0.610\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.299\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.701\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21.2521%;\"\u003e\n \u003cp\u003eReflection4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.3558%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.3443%;\"\u003e\n \u003cp\u003e0.830\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.5091%;\"\u003e\n \u003cp\u003e0.176\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.827%;\"\u003e\n \u003cp\u003e0.824\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eNote: Applied Rotation Method is promax.\u003c/em\u003e\u003cem\u003e\u0026nbsp;F\u003c/em\u003e\u003cem\u003eactor loadings below 0.4 were not displayed.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe Cronbach\u0026apos;s alpha coefficients for the four categories in the 4R Model of Generative AI-Based Emotional Support were as follows: Regulation (0.949), Reinforcement (0.953), Resonance (0.957), and Reflection (0.930). As shown in Table 3, the explained variances for each category were 21.2% for Regulation, 21.0% for Reinforcement, 20.7% for Resonance, and 16.3% for Reflection. The cumulative explained variance was 79.2%, indicating that the four categories together account for a substantial portion of the total variance. These values demonstrate high internal consistency and the meaningful contribution of each dimension to the overall scale.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"678\" class=\"fr-table-selection-hover\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"14\" style=\"width: 16.4892%;\"\u003e\n \u003cp\u003eTable 3. Factor Characteristics\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"3\" style=\"width: 5.4869%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"5\" style=\"width: 1.9007%;\"\u003e\n \u003cp\u003eUnrotated solution\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"6\" style=\"width: 7.3755%;\"\u003e\n \u003cp\u003eRotated solution\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 3.3791%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 2.4706%;\"\u003e\n \u003cp\u003eEigenvalues\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 1.896%;\"\u003e\n \u003cp\u003eSumSq.\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eLoadings\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 13.4218%;\"\u003e\n \u003cp\u003eProportion\u0026nbsp;\u003c/p\u003e\n \u003cp\u003evar.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"\" style=\"width: 15.1918%;\"\u003e\n \u003cp\u003eCumulative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 2.1119%;\"\u003e\n \u003cp\u003eSumSq.\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eLoadings\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2.5668%;\"\u003e\n \u003cp\u003eProportion\u0026nbsp;\u003c/p\u003e\n \u003cp\u003evar.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 2.6643%;\"\u003e\n \u003cp\u003eCumulative\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.3791%;\"\u003e\n \u003cp\u003eRegulation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 0.2274%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 3.2166%;\"\u003e\n \u003cp\u003e14.373\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 1.7545%;\"\u003e\n \u003cp\u003e14.166\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 18.2133%;\"\u003e\n \u003cp\u003e0.708\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 12.0227%;\"\u003e\n \u003cp\u003e0.708\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 0.2599%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 1.5921%;\"\u003e\n \u003cp\u003e4.245\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 2.9567%;\"\u003e\n \u003cp\u003e0.212\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2.3069%;\"\u003e\n \u003cp\u003e0.212\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.3791%;\"\u003e\n \u003cp\u003eReinforcement\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 0.2274%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 3.2166%;\"\u003e\n \u003cp\u003e1.049\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 1.7545%;\"\u003e\n \u003cp\u003e0.840\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 18.2133%;\"\u003e\n \u003cp\u003e0.042\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 12.0227%;\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 0.2599%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 1.5921%;\"\u003e\n \u003cp\u003e4.209\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 2.9567%;\"\u003e\n \u003cp\u003e0.210\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2.3069%;\"\u003e\n \u003cp\u003e0.423\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.3791%;\"\u003e\n \u003cp\u003eResonance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 0.2274%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 3.2166%;\"\u003e\n \u003cp\u003e0.684\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 1.7545%;\"\u003e\n \u003cp\u003e0.477\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 18.2133%;\"\u003e\n \u003cp\u003e0.024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 12.0227%;\"\u003e\n \u003cp\u003e0.774\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 0.2599%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 1.5921%;\"\u003e\n \u003cp\u003e4.135\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 2.9567%;\"\u003e\n \u003cp\u003e0.207\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2.3069%;\"\u003e\n \u003cp\u003e0.629\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.3791%;\"\u003e\n \u003cp\u003eReflection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 0.2274%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 3.2166%;\"\u003e\n \u003cp\u003e0.560\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 1.7545%;\"\u003e\n \u003cp\u003e0.361\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 18.2133%;\"\u003e\n \u003cp\u003e0.018\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 12.0227%;\"\u003e\n \u003cp\u003e0.792\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 0.2599%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 1.5921%;\"\u003e\n \u003cp\u003e3.255\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 2.9567%;\"\u003e\n \u003cp\u003e0.163\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2.3069%;\"\u003e\n \u003cp\u003e0.792\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"14\" style=\"width: 13.4051%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 3.3791%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 0.2274%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 3.2166%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 1.3646%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 0.3899%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11.062%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 4.4965%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 10.9902%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 0.2599%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 0.2599%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 1.5921%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 2.5668%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 0.3574%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 2.3069%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e3.5 Confirmatory factor analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe applied SmartPLS (CB-SEM module) to conduct Confirmatory Factor Analysis (CFA) on the second set of data as shown in Figure 1. CFA was a statistical method used to verify if the model proposed by researchers aligned with the actual data. Because the \u003cem\u003e\u0026chi;2\u003c/em\u003e test statistic was reported, but it was not used as the primary indicator for evaluating model fit. This is because the \u003cem\u003e\u0026chi;2\u003c/em\u003e statistic is known to be sensitive to sample size, and relying on it alone could lead to inappropriate conclusions. Instead, the focus was on other more robust fit indices, such as RMSEA, SRMR, CFI, and TLI, to comprehensively evaluate the fit between the hypothesized model and the observed data. The thresholds for these indicators typically included CFI and TLI greater than 0.90, and SRMR less than 0.08. RMSEA faces considerable difficulties when assessing simpler models with few \u003cem\u003edf\u003c/em\u003e, notably apparent in basic CFAs where \u003cem\u003edf\u003c/em\u003e are often limited. In such instances, RMSEA might erroneously indicate a subpar fit, even when the model accurately represents the data (Kenny et al., 2015). Consequently, we adopt a less strict threshold (0.090) to mitigate this issue.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.6 Model comparison\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConsidering that the partial correlations between the factors were very high, we conducted a model comparison between the four-factor model and the one-factor model. The comparison primarily focused on the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC), where lower values indicate better model fit. In addition, we compared other model fit indices between the four-factor and one-factor models, such as [CFI, TLI, RMSEA, SRMR]. The results in Table 4 showed that the four-factor model outperformed the one-factor model across all fit indices, indicating that the four-factor structure provides a better representation of the data, thereby supporting our research hypothesis.\u003c/p\u003e\n\u003cp\u003eTable 4. Model fit statistics\u003c/p\u003e\n\u003cdiv align=\"center\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"125%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cem\u003eModel\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026chi;2\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 4px;\"\u003e\n \u003cp\u003e\u003cem\u003edf\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026chi;2/df\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003eRMSEA[LOW 90% CI- HIGH 90% CI]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003eSRMR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5px;\"\u003e\n \u003cp\u003eTLI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5px;\"\u003e\n \u003cp\u003eCFI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\n \u003cp\u003eAIC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\n \u003cp\u003eBIC\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e4-factor model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003e693.116\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 4px;\"\u003e\n \u003cp\u003e164\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e4.226\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e0.080\u0026nbsp;[0.074--0.087]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e0.022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5px;\"\u003e\n \u003cp\u003e0.948\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5px;\"\u003e\n \u003cp\u003e0.956\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\n \u003cp\u003e785.116\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\n \u003cp\u003e978.988\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e1-factor model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9px;\"\u003e\n \u003cp\u003e1790.283\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 4px;\"\u003e\n \u003cp\u003e170\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e10.531\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24px;\"\u003e\n \u003cp\u003e0.138 [0.132-0.144]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 6px;\"\u003e\n \u003cp\u003e0.046\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5px;\"\u003e\n \u003cp\u003e0.848\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5px;\"\u003e\n \u003cp\u003e0.864\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\n \u003cp\u003e1870.283\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8px;\"\u003e\n \u003cp\u003e2038.867\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003e3.7 Convergent validity\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe assessment of convergent validity was conducted through several key indicators. First, the standardized factor loadings ranged from 0.823 to 0.915, indicating a strong correlation between the measurement items and their corresponding latent variables, which meets established measurement standards. Second, the CR values fell between 0.937 and 0.956, surpassing the conventional threshold of 0.7, thus demonstrating high internal consistency. Additionally, the AVE values ranged from 0.765 to 0.788, exceeding the acceptable level of 0.5, which suggests that a significant proportion of variance is accounted for by the latent variables. Overall, these results indicate that the scale possesses good convergent validity, making it suitable for further research and analysis. The CR values for all factors fell between 0.867 and 0.944, well above the 0.70 recommended standard, suggesting the scale dimensions have strong internal consistency. The AVE values for the factors ranged from 0.621 to 0.771, all meeting the 0.50 threshold for convergent validity. This implies the scale is able to account for a significant portion of the variance in its constituent items. In summary, the comprehensive analysis of factor loadings, CR, and AVE provides evidence of the strong convergent validity of the scale.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.8 Discriminant validity\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo establish the discriminant validity of the scale, this study utilized the Fornell-Larcker criterion. The Fornell-Larcker criterion stipulates that the square root of the AVE for each construct should be greater than its correlations with other constructs. In the current analysis as displayed in Table 5, we observe that the square root of the AVE for Reflection (0.888) is lower than its correlation with Reinforcement (0.920), and that square root of the AVE for Resonance (0.884) is lower than its correlation with Reinforcement (0.918). This suggests that Reflection and Reinforcement, as well as Resonance and Reinforcement, are not sufficiently distinct, which violates the Fornell-Larcker criterion for discriminant validity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 5. Fornell-Larcker results\u003c/p\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003eReflection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003eRegulation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eReinforcement\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003eResonance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eReflection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.888\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eRegulation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.824\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.874\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eReinforcement\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e\u003cem\u003e0.920\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.843\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.882\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eResonance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.883\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.853\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u003cem\u003e0.918\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.884\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cem\u003eNote: The diagonal values represent the square root of the AVE.\u003c/em\u003e\u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThe primary aim of this study was to develop and validate a scale designed to measure the emotional support provided by generative artificial intelligence (AI). This scale, which consists of four key dimensions\u0026mdash;Regulation, Resonance, Reinforcement, and Reflection\u0026mdash;was created to capture the multifaceted nature of AI-driven emotional support. The development process involved a rigorous validation process, including item analysis, exploratory factor analysis (EFA), and confirmatory factor analysis (CFA).\u003c/p\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003e4.1 Discussion on the rationale of four factors\u003c/h2\u003e\u003cp\u003eThe results from the EFA indicated that the four-factor structure (Regulation, Resonance, Reinforcement, and Reflection) explained a significant portion of the variance in users' perceptions of AI emotional support, with each dimension contributing to the overall model.\u003c/p\u003e\u003cp\u003eThe four dimensions\u0026mdash;Regulation, Resonance, Reinforcement, and Reflection\u0026mdash;are grounded in the capabilities of AI to support emotional well-being. Regulation is well represented by items that focus on AI\u0026rsquo;s ability to help users manage negative emotions, such as through relaxation techniques or providing strategies to cope with stress. These items align with the dimension\u0026rsquo;s focus on emotional control.\u003c/p\u003e\u003cp\u003eResonance is effectively captured by items that highlight AI's capacity to understand, respond to, and validate users' emotions. This emotional connection is key to helping users feel heard and supported, which is central to this dimension.\u003c/p\u003e\u003cp\u003eReinforcement is reflected in items that emphasize AI's role in providing positive feedback, boosting confidence, and motivating users to continue engaging in positive behaviors. This aligns with the definition of reinforcement as encouraging continued progress and self-belief.\u003c/p\u003e\u003cp\u003eFinally, Reflection is represented by items that prompt users to engage in self-reflection about their emotions and behaviors. AI aids users in gaining deeper insights into their emotional responses and behaviors, fostering personal growth, which fits the essence of the Reflection dimension. This alignment between the items and the dimensions they represent supports the accuracy and appropriateness of the chosen labels for each dimension.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003e4.2 Discussion on the overlap between factors\u003c/h2\u003e\u003cp\u003eCFA further supported the findings from EFA, demonstrating a good model fit with the data. Additionally, the convergent validity of the scale was established, as each factor showed high internal consistency. Although the four dimensions showed strong internal consistency and good model fit, some concerns arose regarding the discriminant validity of the scale, particularly between the Reflection and Reinforcement dimensions, as well as between Resonance and Reinforcement. The correlations between these dimensions were higher than expected, indicating that they may not be sufficiently distinct from each other.\u003c/p\u003e\u003cp\u003eFour possible explanations for the inadequate discriminant validity are: overlapping contextual assumptions, interwoven support types, a sample that lacks diversity, and overlap of multiple response strategies. First, although the items in different dimensions belong to different types of support, they are often designed based on similar contextual assumptions. This means that most items assume that generative AI can recognize and respond to users' emotional states. The similarity in these assumptions may lead to overlap between the items in different dimensions during actual use, making it difficult for respondents to distinguish between these types of support, thereby affecting the discriminant validity between dimensions.\u003c/p\u003e\u003cp\u003eSecond, these types of support are often not independent. For example, users may experience both emotional regulation and emotional reinforcement during their interaction with AI. In the process of emotional regulation, AI may further enhance the user's emotional state through positive feedback. This interweaving and interaction between different support types may blur the boundaries between dimensions, thus affecting their discriminant validity.\u003c/p\u003e\u003cp\u003eThirdly, the homogeneity of the sample may also impact discriminant validity. If the characteristics of the sample are too concentrated, it could affect the diversity of responses when answering items related to different dimensions. Such homogeneity in the sample may lead to more consistent responses to certain dimensions, reducing the ability to distinguish between them.\u003c/p\u003e\u003cp\u003eFourth, generative AI tools often employ multiple approaches simultaneously when generating responses. These tools combine various algorithms and strategies, which can result in overlapping or intertwined emotional expressions across different dimensions. This multi-faceted response mechanism can blur the boundaries between dimensions, making it difficult to clearly distinguish between them. Therefore, while these dimensions are conceptually distinct, their practical application in generative AI responses may lead to interactions between them that reduce their differentiation.\u003c/p\u003e\u003c/div\u003e"},{"header":"5. Limitations and suggestions","content":"\u003cp\u003eAlthough this study provides valuable insights, there are several limitations that should be considered. First, although the sample size is sufficient, all the samples are from the same school. Despite the inclusion of students from different academic backgrounds, there may still be issues of homogeneity, which could affect the generalizability of the research findings. Therefore, future studies could consider collecting data from multiple schools or different regions to increase sample diversity and improve the external validity of the findings.\u003c/p\u003e\u003cp\u003eSecond, this study adopts a cross-sectional design, which means that repeated measurements from the same participants were not collected, and test-retest reliability was not assessed. Therefore, future research could employ a longitudinal design to track changes in participants over time, in order to verify the stability and reliability of the results.\u003c/p\u003e\u003cp\u003eThird, this study relied solely on Classical Test Theory for measurement. Future research could incorporate other methodologies, such as network analysis or Item Response Theory (IRT), to provide a more comprehensive validation of the findings. These additional methods could help improve the robustness and depth of the measurement process, offering further insights into the reliability and validity of the results.\u003c/p\u003e\u003cp\u003eFinally, this research is specifically focused on Artificial Intelligence Generated Content technology in a general context, without targeting any particular product, e.g., Doubao or ChatGPT. Although the findings shed light on the potential emotional support capabilities of AIGC, they may not be fully applicable to specific applications. Therefore, future research can validate the generalizability of the findings by testing AIGC in different products, examining its effectiveness and emotional support capabilities across various real-world contexts.\u003c/p\u003e"},{"header":"6. Implications","content":"\u003cp\u003eGiven the rising number of mental health issues and the increasing demand for counseling services, the potential of AI in providing emotional support has become more apparent. As more people begin to rely on AI technology for emotional support, it has become essential to measure individuals' perceptions of the emotional support provided by generative AI. The findings of this study have several important implications for both research and practice.\u003c/p\u003e\u003cp\u003eFirst, in the field of psychological counseling, the study highlights how generative artificial intelligence can play a supportive role in emotional regulation, offering an innovative tool to enhance mental health interventions. By integrating AI-based emotional support into counseling sessions, mental health professionals can provide more personalized and responsive care.\u003c/p\u003e\u003cp\u003eSecond, this study emphasizes how developers can further enhance these features to make emotional support more targeted and effective. By incorporating more advanced emotional intelligence into AI systems, developers can ensure that the AI tools are better at recognizing, resonating with, and responding to users' emotional states. This allows for more personalized and empathetic interactions, improving the user experience and satisfaction.\u003c/p\u003e\u003cp\u003eThird, from a practical perspective, by leveraging generative AI to provide immediate emotional responses and guidance, these services can be scaled up, reducing the need for extensive human intervention. This approach makes emotional support more affordable and accessible to a wider range of individuals. In addition, the integration of AI can help protect user privacy by offering emotional support without the need for face-to-face interactions. This could reduce the stigma often associated with seeking psychological help, allowing individuals to access support without the fear of being labeled or judged. By providing anonymous assistance, AI helps maintain privacy while delivering much-needed emotional guidance.\u003c/p\u003e"},{"header":"7. Conclusion","content":"\u003cp\u003eIn conclusion, this study highlights the significant potential of AIGC technology in providing emotional support across various contexts. By exploring its capabilities in emotional regulation, resonance, reinforcement, and reflection, the research offers valuable insights into how AIGC can be used to address the emotional needs of individuals. Despite the limitations, this research paves the way for a deeper exploration of AIGC's integration into emotionally intelligent systems, offering new possibilities for innovation in AI-driven solutions.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Macau Millennium College Research Ethics Committee (Approval No.: MMCIRB-2024-001). A digital consent form was embedded in the questionnaire, and participants could only proceed after clicking the \u0026quot;Agree to Participate\u0026quot; button. In addition, the participants were fully informed about the study\u0026apos;s objectives, procedures, risks, and benefits prior to giving their consent. By clicking the \u0026quot;Agree to Participate\u0026quot; button, they confirmed their understanding and voluntary agreement to take part in the research. The process ensured that all participants provided informed consent electronically before engaging with the survey. All methods were performed in accordance with the Declaration of Helsinki.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe embedded digital consent form specified the purpose of the study, and all research data are solely used for academic purposes. All participants were informed that the results of the study may be published in academic journals or presented at conferences.\u003c/p\u003e\n\u003ch2\u003eCompeting interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eYYC: Responsible for research design, data collection, and analysis.YXX and LXY: Responsible for project supervision and final manuscript preparation.WQ: Responsible for experimental implementation and data management.WXF: Responsible for manuscript editing.TZH: Responsible for literature review, data analysis, results discussion, and initial draft writing.\u003c/p\u003e\n\u003ch2\u003eAcknowledgement\u003c/h2\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe datasets used and/or analysed during the current study available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAgnafors, S., Barmark, M. \u0026amp; Sydsj\u0026ouml;, G. Mental health and academic performance: a study on selection and causation effects from childhood to early adulthood. \u003cem\u003eSoc. Psychiatry Psychiatr. Epidemiol.\u003c/em\u003e \u003cb\u003e56\u003c/b\u003e, 857\u0026ndash;866 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eArifudin, M., Robbaniyah, I. \u0026amp; Ramadhan, I. Confiding through Artificial Intelligence: Can ChatGPT Provide Emotional Support? \u003cem\u003eProceeding of the International Conference on Social Sciences and Humanities Innovation\u003c/em\u003e, \u003cem\u003e1\u003c/em\u003e(1), 69\u0026ndash;78. (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAult, S., Breitenstein, S. M., Tucker, S., Havercamp, S. M. \u0026amp; Ford, J. L. Caregivers of children with autism spectrum disorder in rural areas: A literature review of mental health and social support. \u003cem\u003eJ. Pediatr. Nurs.\u003c/em\u003e \u003cb\u003e61\u003c/b\u003e, 229\u0026ndash;239 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBarratt, J. M. \u0026amp; Duran, F. Does psychological capital and social support impact engagement and burnout in online distance learning students? \u003cem\u003eInternet High. Educ.\u003c/em\u003e \u003cb\u003e51\u003c/b\u003e, 100821 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBearman, M. \u0026amp; Ajjawi, R. Learning to work with the black box: Pedagogy for a world with artificial intelligence. \u003cem\u003eBr. J. Edu. Technol.\u003c/em\u003e \u003cb\u003e54\u003c/b\u003e (5), 1160\u0026ndash;1173 (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBedaso, A., Adams, J., Peng, W. \u0026amp; Sibbritt, D. The relationship between social support and mental health problems during pregnancy: a systematic review and meta-analysis. \u003cem\u003eReproductive Health\u003c/em\u003e. \u003cb\u003e18\u003c/b\u003e, 1\u0026ndash;23 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBuck, D. et al. M., \u003cem\u003eImpact of emotional support for mental health student nurses and retention on the mental health nursing programme.\u003c/em\u003e [Vrije Universiteit Amsterdam]. (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5463/thesis.859\u003c/span\u003e\u003cspan address=\"10.5463/thesis.859\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCheng Lin, K., Huang, T. C., Hung, J. C., Yen, N. Y., Chen, J. \u0026amp; S Facial emotion recognition towards affective computing-based learning. \u003cem\u003eLibr. Hi Tech.\u003c/em\u003e \u003cb\u003e31\u003c/b\u003e (2), 294\u0026ndash;307. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1108/07378831311329068\u003c/span\u003e\u003cspan address=\"10.1108/07378831311329068\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChu, T., Liu, X., Takayanagi, S., Matsushita, T. \u0026amp; Kishimoto, H. Association between mental health and academic performance among university undergraduates: The interacting role of lifestyle behaviors. \u003cem\u003eInternational Journal of Methods in Psychiatric Research\u003c/em\u003e, \u003cem\u003e32\u003c/em\u003e(1), e1938. (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eD\u0026iacute;az, Y., Hessel, P., Avendano, M. \u0026amp; Evans-Lacko, S. Multidimensional poverty and adolescent mental health: Unpacking the relationship. \u003cem\u003eSoc. Sci. Med.\u003c/em\u003e \u003cb\u003e311\u003c/b\u003e, 115324 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDuncan, M. J., Patte, K. A. \u0026amp; Leatherdale, S. T. Mental health associations with academic performance and education behaviors in Canadian secondary school students. \u003cem\u003eCan. J. School Psychol.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e (4), 335\u0026ndash;357 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEtori, N. A. \u0026amp; Gini, M. WisCompanion: Integrating the Socratic Method with ChatGPT-Based AI for Enhanced Explainability in Emotional Support for Older Adults. \u003cem\u003eInternational Conference on Human-Computer Interaction\u003c/em\u003e, 179\u0026ndash;198. (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFornell, C. \u0026amp; Larcker, D. F. Evaluating structural equation models with unobservable variables and measurement error. \u003cem\u003eJ. Mark. Res.\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e (1), 39\u0026ndash;50. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2307/3151312\u003c/span\u003e\u003cspan address=\"10.2307/3151312\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1981).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGaspar, T., Cerqueira, A., Guedes, F. B. \u0026amp; de Matos, M. G. Parental Emotional Support, Family Functioning and Children\u0026rsquo;s Quality of Life. \u003cem\u003ePsychol. Stud.\u003c/em\u003e \u003cb\u003e67\u003c/b\u003e (2), 189\u0026ndash;199 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGibson, B., Schneider, J., Talamonti, D. \u0026amp; Forshaw, M. The impact of inequality on mental health outcomes during the COVID-19 pandemic: A systematic review. \u003cem\u003eCan. Psychol.\u003c/em\u003e \u003cb\u003e62\u003c/b\u003e (1), 101 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGilbert, M. H., Dextras-Gauthier, J., St-Hilaire, F. \u0026amp; Lerouge, L. Understanding managers\u0026rsquo; mental health: the cornerstone for better organizational performance and workers\u0026rsquo; health. In Frontiers in Psychology (Vol. 15, 1468825). Frontiers Media SA. (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoel, N. J., Thomas, B., Boutt\u0026eacute;, R. L., Kaur, B. \u0026amp; Mazzeo, S. E. What will people say? Mental health stigmatization as a barrier to eating disorder treatment-seeking for South Asian American women. \u003cem\u003eAsian Am. J. Psychol.\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e (1), 96 (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuo, K. et al. Assessing social support impact on depression, anxiety, and stress among undergraduate students in Shaanxi province during the COVID-19 pandemic of China. \u003cem\u003ePloS One\u003c/em\u003e. \u003cb\u003e16\u003c/b\u003e (7), e0253891 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHair, J., Black, W., Babin, B. \u0026amp; Anderson, R. \u003cem\u003eMultivariate data analysis\u003c/em\u003e (7th Editio). Prentice-Hall, Inc. (2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHassija, V. et al. Interpreting black-box models: a review on explainable artificial intelligence. \u003cem\u003eCogn. Comput.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e (1), 45\u0026ndash;74 (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHerbener, A. B. \u0026amp; Damholdt, M. F. Are lonely youngsters turning to chatbots for companionship? The relationship between chatbot usage and social connectedness in Danish high-school students. \u003cem\u003eInt. J. Hum. Comput. Stud.\u003c/em\u003e \u003cb\u003e196\u003c/b\u003e, 103409 (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHinkson, K. D. Jr et al. An examination of the mental health and academic performance of student veterans. \u003cem\u003eJ. Am. Coll. Health\u003c/em\u003e. \u003cb\u003e70\u003c/b\u003e (8), 2334\u0026ndash;2341 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHolliman, A. J. et al. Adaptability and social support: Examining links with psychological wellbeing among UK students and non-students. \u003cem\u003eFront. Psychol.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 636520 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHouse, J. S. \u003cem\u003eWork stress and social support\u003c/em\u003e (Addison-Wesley Series on Occupational Stress, 1983).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHu, L. T. \u0026amp; Bentler, P. M. Cutoff Criteria for Fit Indexes in Covariance Structure Analysis: Conventional Criteria Versus New Alternatives. \u003cem\u003eStruct. Equation Modeling: Multidisciplinary J.\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e (1), 1\u0026ndash;55. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10705519909540118\u003c/span\u003e\u003cspan address=\"10.1080/10705519909540118\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1999).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHu, X. et al. Understanding the impact of emotional support on mental health resilience of the community in the social media in Covid-19 pandemic. \u003cem\u003eJ. Affect. Disord.\u003c/em\u003e \u003cb\u003e308\u003c/b\u003e, 360\u0026ndash;368 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang, L. \u0026amp; Zhang, T. Perceived social support, psychological capital, and subjective well-being among college students in the context of online learning during the COVID-19 pandemic. \u003cem\u003eAsia-Pacific Educ. Researcher\u003c/em\u003e. \u003cb\u003e31\u003c/b\u003e (5), 563\u0026ndash;574 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang, Y. et al. The impacts of coping style and perceived social support on the mental health of undergraduate students during the early phases of the COVID-19 pandemic in China: a multicenter survey. \u003cem\u003eBMC Psychiatry\u003c/em\u003e. \u003cb\u003e21\u003c/b\u003e, 1\u0026ndash;12 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJauch, M., Occhipinti, S. \u0026amp; O\u0026rsquo;Donovan, A. The stigmatization of mental illness by mental health professionals: Scoping review and bibliometric analysis. \u003cem\u003ePLoS One\u003c/em\u003e, \u003cb\u003e18\u003c/b\u003e(1), e0280739. (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKaiser, H. F. An index of factorial simplicity. \u003cem\u003ePsychometrika\u003c/em\u003e \u003cb\u003e39\u003c/b\u003e (1), 31\u0026ndash;36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/BF02291575\u003c/span\u003e\u003cspan address=\"10.1007/BF02291575\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1974).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKaradaş, A. \u0026amp; Duran, S. The effect of social support on work stress in health workers during the pandemic: The mediation role of resilience. \u003cem\u003eJ. Community Psychol.\u003c/em\u003e \u003cb\u003e50\u003c/b\u003e (3), 1640\u0026ndash;1649 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKenny, D. A., Kaniskan, B. \u0026amp; McCoach, D. B. The Performance of RMSEA in Models With Small Degrees of Freedom. \u003cem\u003eSociol. Methods Res.\u003c/em\u003e \u003cb\u003e44\u003c/b\u003e (3), 486\u0026ndash;507. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0049124114543236\u003c/span\u003e\u003cspan address=\"10.1177/0049124114543236\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKikas, E. \u0026amp; Tang, X. Child-perceived teacher emotional support, its relations with teaching practices, and task persistence. \u003cem\u003eEur. J. Psychol. Educ.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e, 359\u0026ndash;374 (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi, W. et al. Socioeconomic inequality in child mental health during the COVID-19 pandemic: First evidence from China. \u003cem\u003eJ. Affect. Disord.\u003c/em\u003e \u003cb\u003e287\u003c/b\u003e, 8\u0026ndash;14 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi, X. Y. The relationship between quality of sports friendships and mental health in Chinese junior high school students: the bidirectional chain mediating effects of sport motivation and exercise adherence. \u003cem\u003eBMC Public. Health\u003c/em\u003e. \u003cb\u003e25\u003c/b\u003e (1), 52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12889-025-21287-5\u003c/span\u003e\u003cspan address=\"10.1186/s12889-025-21287-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLoiacono, E. T., Watson, R. T. \u0026amp; Goodhue, D. L. WebQual: A measure of website quality. \u003cem\u003eMark. Theory Appl.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e (3), 432\u0026ndash;438 (2002).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMai, Y., Wu, Y. J. \u0026amp; Huang, Y. What type of social support is important for student resilience during COVID-19? A latent profile analysis. \u003cem\u003eFront. Psychol.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 646145 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMcDonald, R. P. \u0026amp; Ho, M. H. R. Principles and practice in reporting structural equation analyses. \u003cem\u003ePsychol. Methods\u003c/em\u003e. \u003cb\u003e7\u003c/b\u003e (1), 64\u0026ndash;82. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/1082-989X.7.1.64\u003c/span\u003e\u003cspan address=\"10.1037/1082-989X.7.1.64\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2002).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMendoza-Pinto, R. Artificial Intelligence in the Fight Against Bullying: Integration of ChatGPT in an Emotional Support Chatbot. \u003cem\u003eProceedings Http://Ceur-Ws. Org ISSN\u003c/em\u003e, \u003cem\u003e1613\u003c/em\u003e, 73. (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoore, P. Do We Understand the Relationship between Affective Computing, Emotion and Context-Awareness? \u003cem\u003eMachines (Basel)\u003c/em\u003e. \u003cb\u003e5\u003c/b\u003e (3), 16. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/machines5030016\u003c/span\u003e\u003cspan address=\"10.3390/machines5030016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNaher, J. Can ChatGPT provide a better support: a comparative analysis of ChatGPT and dataset responses in mental health dialogues. \u003cem\u003eCurrent Psychology\u003c/em\u003e, 1\u0026ndash;9. (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eO\u0026rsquo;Connor, R. C. et al. Mental health and well-being during the COVID-19 pandemic: longitudinal analyses of adults in the UK COVID-19 Mental Health \u0026amp; Wellbeing study. \u003cem\u003eBr. J. Psychiatry\u003c/em\u003e. \u003cb\u003e218\u003c/b\u003e (6), 326\u0026ndash;333 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePei, G. et al. Affective Computing: Recent Advances, Challenges, and Future Trends. \u003cem\u003eIntelligent Computing\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e. (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.34133/icomputing.0076\u003c/span\u003e\u003cspan address=\"10.34133/icomputing.0076\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePermatasari, N., Ashari, F. R. \u0026amp; Ismail, N. Contribution of perceived social support (peer, family, and teacher) to academic resilience during COVID-19. \u003cem\u003eGold. Ratio Social Sci. Educ.\u003c/em\u003e \u003cb\u003e1\u003c/b\u003e (1), 1\u0026ndash;12 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePicard, R. W. \u003cem\u003eAffective computing\u003c/em\u003e (MIT Press, 2000).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePolanin, J. R. et al. A meta-analysis of longitudinal partial correlations between school violence and mental health, school performance, and criminal or delinquent acts. \u003cem\u003ePsychol. Bull.\u003c/em\u003e \u003cb\u003e147\u003c/b\u003e (2), 115 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRashid Soron, T. et al. H., C Domestic violence and mental health during the COVID-19 pandemic in Bangladesh. \u003cem\u003eJMIR Formative Res.\u003c/em\u003e, \u003cb\u003e5\u003c/b\u003e(9), e24624. (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRider, E. A., Ansari, E., Varrin, P. H. \u0026amp; Sparrow, J. Mental health and wellbeing of children and adolescents during the covid-19 pandemic. \u003cem\u003eBmj\u003c/em\u003e, \u003cem\u003e374\u003c/em\u003e. (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRohan, R., Faruk, L. I. D., Puapholthep, K. \u0026amp; Pal, D. Unlocking the Black Box: Exploring the use of Generative AI (ChatGPT) in Information Systems Research. \u003cem\u003eProceedings of the 13th International Conference on Advances in Information Technology\u003c/em\u003e, 1\u0026ndash;9. (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRomm, K. F., Metzger, A. \u0026amp; Turiano, N. A. Parental emotional support and health problems: The role of social support and social strain. \u003cem\u003eJ. Adult Dev.\u003c/em\u003e \u003cb\u003e28\u003c/b\u003e (4), 319\u0026ndash;331 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchenke, K., Ruzek, E., Lam, A. C., Karabenick, S. A. \u0026amp; Eccles, J. S. To the means and beyond: Understanding variation in students\u0026rsquo; perceptions of teacher emotional support. \u003cem\u003eLearn. Instruction\u003c/em\u003e. \u003cb\u003e55\u003c/b\u003e, 13\u0026ndash;21 (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShin, D. et al. Exploring the effects of AI-assisted emotional support processes in online mental health community. \u003cem\u003eCHI Conference on Human Factors in Computing Systems Extended Abstracts\u003c/em\u003e, 1\u0026ndash;7. (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShuo, Z., Xuyang, D., Xin, Z., Xuebin, C. \u0026amp; Jie, H. The relationship between postgraduates\u0026rsquo; emotional intelligence and well-being: the chain mediating effect of social support and psychological resilience. \u003cem\u003eFront. Psychol.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 865025 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSteiger, S. et al. Personality, self-esteem, familiarity, and mental health stigmatization: a cross-sectional vignette-based study. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e (1), 10347 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eThomson, R. M., Kopasker, D., Leyland, A., Pearce, A. \u0026amp; Katikireddi, S. V. Effects of poverty on mental health in the UK working-age population: causal analyses of the UK Household Longitudinal Study. \u003cem\u003eInt. J. Epidemiol.\u003c/em\u003e \u003cb\u003e52\u003c/b\u003e (2), 512\u0026ndash;522 (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTibber, M. S., Walji, F., Kirkbride, J. B. \u0026amp; Huddy, V. The association between income inequality and adult mental health at the subnational level\u0026mdash;a systematic review. \u003cem\u003eSocial Psychiatry Psychiatric Epidemiology\u003c/em\u003e, 1\u0026ndash;24. (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, Y. et al. A systematic review on affective computing: emotion models, databases, and recent advances. \u003cem\u003eInform. Fusion\u003c/em\u003e. \u003cb\u003e83\u0026ndash;84\u003c/b\u003e, 19\u0026ndash;52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.inffus.2022.03.009\u003c/span\u003e\u003cspan address=\"10.1016/j.inffus.2022.03.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWHO. \u003cem\u003eMental health\u003c/em\u003e. World Health Organization. (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/news-room/fact-sheets/detail/mental-health-strengthening-our-response\u003c/span\u003e\u003cspan address=\"https://www.who.int/news-room/fact-sheets/detail/mental-health-strengthening-our-response\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWolfinbarger, M. \u0026amp; Gilly, M. C. eTailQ: dimensionalizing, measuring and predicting etail quality. \u003cem\u003eJ. Retail.\u003c/em\u003e \u003cb\u003e79\u003c/b\u003e (3), 183\u0026ndash;198 (2003).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYeung, R. \u0026amp; Leadbeater, B. Adults make a difference: the protective effects of parent and teacher emotional support on emotional and behavioral problems of peer-victimized adolescents. \u003cem\u003eJ. Community Psychol.\u003c/em\u003e \u003cb\u003e38\u003c/b\u003e (1), 80\u0026ndash;98 (2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZaneva, M., Guzman-Holst, C., Reeves, A. \u0026amp; Bowes, L. The impact of monetary poverty alleviation programs on children\u0026rsquo;s and adolescents\u0026rsquo; mental health: A systematic review and meta-analysis across low-, middle-, and high-income countries. \u003cem\u003eJ. Adolesc. Health\u003c/em\u003e. \u003cb\u003e71\u003c/b\u003e (2), 147\u0026ndash;156 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZeng, Z. \u003cem\u003eEmotional Support and Mental Health: Investigating Interpersonal Emotion Regulation in Chinese and Dutch Contexts\u003c/em\u003e. (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou, Z. \u0026amp; Cheng, Q. Relationship between online social support and adolescents\u0026rsquo; mental health: A systematic review and meta-analysis. \u003cem\u003eJ. Adolesc.\u003c/em\u003e \u003cb\u003e94\u003c/b\u003e (3), 281\u0026ndash;292 (2022).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"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":"Generative Artificial Intelligence, Emotional Support, Regulation, Resonance, Reinforcement, Reflection","lastPublishedDoi":"10.21203/rs.3.rs-6956402/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6956402/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study conceptualizes and validates the 4R Model of Generative AI-Based Emotional Support Scale, encompassing four key dimensions: Regulation (5 items), Resonance (6 items), Reinforcement (5 items), and Reflection (4 items). These dimensions were developed based on expert interviews and are intended to assess the effectiveness of generative AI in providing emotional support. The study sample consisted of 996 participants from a northern Chinese university, with a convenience sampling method used to collect data. The validation process followed multiple steps, including item analysis, exploratory factor analysis, and confirmatory factor analysis, assessing its reliability, content validity, structural validity, convergent validity and discriminant validity. The findings indicate that the scale demonstrates satisfactory reliability and construct validity. However, while overall construct validity is acceptable, the scale exhibited some challenges in discriminant validity. This limitation can be attributed to several factors, including overlapping contextual assumptions, the interwoven nature of different support types, the lack of diversity in the sample, and the overlap of multiple response strategies. These factors likely contributed to the observed difficulties in achieving strong discriminant validity. Key limitations of the study include concerns regarding the representativeness of the sample, the cross-sectional design that precluded the assessment of test-retest reliability, the reliance on a single validation method, and the focus on generative AI technology in general rather than specific AI-based products. Corresponding solutions are proposed, including expanding the sample to ensure greater diversity, employing a longitudinal design to assess test-retest reliability, incorporating multiple validation methods, and focusing on specific AI-based products in future research. In conclusion, while the 4R Model offers a robust framework for evaluating generative AI-based emotional support, further research is necessary to refine the scale, particularly in terms of improving discriminant validity and expanding its applicability across more diverse contexts.\u003c/p\u003e","manuscriptTitle":"Conceptualizing and Validating the 4R Model of Generative AI-Based Emotional Support Scale: Regulation, Resonance, Reinforcement, and Reflection","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-25 11:39:44","doi":"10.21203/rs.3.rs-6956402/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":"385d87f0-2987-4654-bba9-2837e1422c7f","owner":[],"postedDate":"July 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":51935136,"name":"Physical sciences/Mathematics and computing/Information technology"},{"id":51935137,"name":"Biological sciences/Psychology"}],"tags":[],"updatedAt":"2025-09-25T09:39:05+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-25 11:39:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6956402","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6956402","identity":"rs-6956402","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}