Proposal of INDIRECT X Mentions as an Altmetrics Indicator: Dissemination of Research Papers on X via Web News and Blogs

Proposal of INDIRECT X Mentions as an Altmetrics Indicator: Dissemination of Research Papers on X via Web News and Blogs | 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 Research Article Proposal of INDIRECT X Mentions as an Altmetrics Indicator: Dissemination of Research Papers on X via Web News and Blogs Ai Kishimoto, Takayuki Hayashi This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8121384/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 reexamines the potential of altmetrics as indicators for evaluating the societal impact of research, focusing specifically on X (formerly Twitter). Existing X-based metrics primarily measure first-order citations —posts that directly link to academic papers. However, such metrics overlook second-order citations , which disseminate research indirectly through intermediary webpages such as news articles and blogs. To address this gap, we propose a new indicator, INDIRECT X mentions, designed to quantify the indirect dissemination of scholarly outputs. Using data from the artificial intelligence (AI) domain, this study compares the performance of DIRECT and INDIRECT X mentions over a 1.5-year period following publication. The results reveal that second-order citations occur at a comparable scale to first-order citations and exhibit distinctive diffusion characteristics: while INDIRECT X mentions demonstrate lower immediacy, they sustain engagement over a longer period. Furthermore, network analysis shows that the two metrics are generated by largely non-overlapping user groups, with INDIRECT X mentions associated more strongly with non-academic and media users. These findings suggest that INDIRECT X mentions provide a complementary perspective for measuring how academic knowledge circulates and resonates beyond scholarly communities, offering a more comprehensive framework for assessing the societal reach of research. Altmetrics INDIRECT X mensions Twitter social impact second-order citations Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction The evaluation of academic research has traditionally been based on citation counts and citation-based metrics. These indicators have been widely used not only in academia but also in policymaking. However, concerns have been raised regarding their reliability (Gingras, 2014 ; van Raan, 2019 ; Wouters, 2014 ), time lag before citation counts accumulate (Priem, 2014 ; Brody & Harnad, 2006) and inability to measure the societal impact of research beyond the academic community (Bornmann & Haunschild, 2019 ; Priem, 2014 ). Since the 2000s, the widespread use of the Internet has led to an increase in the need for new evaluation metrics, and Priem et al. ( 2010 ) proposed the concept of "altmetrics". Currently, the practical application of altmetric indicators has been achieved by multiple data providers, most notably Altmetric.com 3 , PlumX 4 and Impactstory 5 . As of 2022, Altmetric.com had successfully tracked more than 200 million mentions associated with approximately 20 million research outputs, thereby demonstrating its capability to visualize the societal impact of scholarly papers in real time that cannot be measured solely by citation counts 6 (Condon,2022). Among altmetrics data sources, X (formerly Twitter) has drawn particular attention because of its data scale (Alhoori & Furuta, 2014 ) and potential as an indicator for measuring the societal impact of research (Wonters et al., 2019). Notably, academic discussions on X have been increasing, and further enhancing its value as a data source (Yu et al., 2019 ). Priem and Costello ( 2010 ) stated that the dissemination of academic papers on X involves first-order citations, which hyperlink directly to academic resources, and second-order citations, which hyperlink to intermediary webpages such as news articles or blogs that describe and link to the resources. Currently, both the X-based indicators provided by data providers 7 and many previous studies tend to focus solely on first-order citations. However, it has been shown that users who post first-order citations differ from the general population, exhibiting a bias toward individuals with academic backgrounds (Haustein, 2019 ; Costas, 2020; Zhang, 2023). Conversely, several prior studies have suggested that second-order citations may exist on a comparable or even larger scale than first-order citations (Priem & Costello, 2010 , Ke, at al., 2017). A recent case study by Alperin et al. ( 2024 ) examined COVID-19 research by focusing on five prominent English-language news outlets— BBC, MSN, The New York Times, The Guardian, and The Washington Post. They analyzed the second-order citations generated by news articles published in these outlets and found that such citations involved approximately twice as many unique posting users and engagement counts as first-order citations. Moreover, the user groups associated with first- and second-order citations showed minimal overlap, with the latter group containing a markedly lower proportion of researchers. These findings illustrate how scientific knowledge was disseminated to the broader public in the context of COVID-19, where public interest in science was particularly high. However, in more general research domains, knowledge is likely to be disseminated not only through major news outlets but also via a wider range of online media such as specialized news sites and blogs. Therefore, it is essential to evidence from diverse research domains to clarify how second-order citations differ from first-order citations in terms of frequency, user groups, and diffusion dynamics. In this study, we focus on second-order citations and analyze them using a newly defined indicator, which we refer to as INDIRECT X mentions, that encompasses a broader range of online media including diverse news sites and blogs. To clearly distinguish between the two indicators, we define DIRECT X mentions as a conventional metric that captures the direct dissemination of research papers through academic paper websites via first-order citations, and INDIRECT X mentions as a new metric that captures indirect dissemination through intermediary webpages via second-order citations. As a case study, this research focuses on the field of artificial intelligence (AI). In June 2018, OpenAI introduced GPT, and on November 30, 2022, the organization released ChatGPT 8 — a web-based conversational service powered by GPT-3.5— to the general public. The system rapidly gained unprecedented attention, surpassing 100 million users within just two months of its release. This breakthrough marked a transformative moment that propelled artificial intelligence (AI) into a globally significant research domain attracting broad interest not only from the academic community but also from industry and the general public. Despite this surge in societal attention, the social impact of AI-related research on X has not been sufficiently examined. To address this gap, the present study focuses on academic papers published between November 2022—the month ChatGPT was released—and April 2023. From this set, we selected 100 AI-related papers with the highest DIRECT X mentions and conducted a comparative analysis of two indicators—DIRECT and INDIRECT X mentions—over a 1.5-year period following publication. Although these papers are typically regarded as highly influential according to conventional altmetric rankings, this study demonstrates that incorporating INDIRECT X mentions can lead to a substantially different evaluation of their impact. By addressing the following research questions, we aim to clarify the existence and distinct diffusion patterns of paper dissemination via second-order citations that are not captured by first-order citations, emphasize the importance of measuring such second-order dissemination on X, and contribute to improving the validity and comprehensiveness of altmetric indicators. RQ1. What kinds of dissemination patterns can be identified when INDIRECT X mentions are incorporated into existing metrics (Score Aggregation Analysis)? RQ2. How do the temporal dynamics of paper dissemination differ between first-order and second-order citations (Time-Series Analysis)? RQ3. How does the user network differ between DIRECT and INDIRECT X mentions (Network Analysis)? 2. Method and data 2.1 Method for Collecting INDIRECT X mentions The existing indicator, DIRECT X mentions, is defined as the sum of the total number of posts that directly hyperlink to the webpage of an academic paper (first-order citations) and their dissemination counts, including reposts and quotes. To enable comparative analysis, the new indicator proposed in this study, INDIRECT X mentions, is constructed in a manner similar to the existing metric. It sums the total number of posts that contain hyperlinks to intermediary webpages referring to academic papers (second-order citations) and their corresponding dissemination counts, including reposts and quotes. The aggregation of second-order citations was conducted using a two-step process. First, intermediary webpage data were collected using the “Mentions” category within the Altmetrics database, which aggregates News and Blogs data. Second, posts on X that included hyperlinks to these intermediary webpages were retrieved via the X API, and their reposts and quotes were counted. By aggregating these data, the INDIRECT X mentions score was systematically quantified. Through this method, the study establishes a consistent and comprehensive framework for measuring INDIRECT X mentions, enabling the systematic quantification of second-order citation-based dissemination. 2.2 Data on Papers, DIRECT X mentions, and INDIRECT X mentions In this study, we used data provided by Altmetric.com to select the target papers and to obtain both the DIRECT X mentions scores and the intermediary webpage data. Although several data providers supply altmetric data, their data collection methodologies and coverage vary considerably. Among them, Altmetric.com is known for its comprehensive coverage of X (formerly Twitter) data and intermediary webpage data (Ortega, J. L., 2018 ; Ortega, J. L., 2020a ; Zahedi et al., 2018). Since the X score provided by Altmetric.com corresponds to the DIRECT X mentions defined in this study—which include only first-order citations—we used Altmetric.com’s existing X metrics as the basis for paper selection. This study focuses on the field of artificial intelligence (AI). Unlike COVID-19 by Alperin et al. ( 2024 ), which represented an exceptional crisis situation with extraordinarily high public urgency, AI represents a more typical yet increasingly prominent research domain that attracts substantial interest from diverse audiences including the general public, technology practitioners, policymakers, and researchers. This makes AI particularly suitable for examining the dissemination patterns of scientific knowledge under more generalizable conditions. Target papers in the AI field were identified using the Altmetric.com subject codes (4602, 4611) and a keyword search applied to titles, including AI, Artificial Intelligence, Deep Learning, GPT, LLM , and Large Language Model . The publication period was restricted to papers published between November 1, 2022, and April 30, 2023. For papers available on arXiv, the date of appearance on arXiv was treated as the official publication date because these preprints are often publicly disseminated and cited well before formal journal publication, thereby representing the earliest point of online visibility. From the collected results, the top 100 papers were selected based on the DIRECT X mentions values measured 1.5 years after publication. Subsequently, for these 100 papers, we obtained daily time-series data of DIRECT X mentions and a list of intermediary webpages (URLs of Web News and Blogs) from Altmetric.com. As a result, 4,473 intermediary webpages were collected. For second-order citation data, we used NodeXL 9 , a social network analysis tool for collecting and visualizing social media data, to collect posts on X that contained URLs linking to the intermediary webpages. These collected posts numbered 28,926 in total. Based on these data, we aggregated the total number of posts, reposts, and quotes to calculate the INDIRECT X mentions scores for the 100 target papers. 2.3 Data for Network Analysis The dataset for the network analysis included only posts containing relevant links and excluded its reshare posts. Each post consisted of a posting user and one or more mentioned users, forming a user interaction network. To clarify communication patterns between users, posts in which the posting user and the mentioned user were identical (such as self-replies or self-quotes) were excluded from the analysis. Usernames were converted to lowercase strings to ensure consistency and to check for duplicates. For analytical purposes, the network formed by DIRECT X mentions was defined as Group 1, and that formed by INDIRECT X mentions was defined as Group 2. Due to data access limitations and significant cost constraints of the X API—particularly for retrieving posts linking to intermediary webpages and their associated metadata—a subset of papers was selected for this analysis. To better capture the characteristics of papers classified into the four quadrants in the RQ1 analysis, In total, 17 papers were included as the target dataset. While the resulting sample is relatively small, it was strategically selected to ensure diversity in dissemination patterns, as explained later. For Group 1 (first-order citations), data were recollected using the X API based on the tweet_id values provided by Altmetric.com. From the aggregated data, 20,328 tweet_ids were identified. However, due to privacy settings and account deletions, detailed post data could be retrieved for 20,107 posts via the X API. After filtering for posts within the observation period, 5,090 posts remained as the final dataset for analysis. For Group 2 (second-order citations), data were extracted from the collection used in the INDIRECT X mentions aggregation, restricted to the selected papers, resulting in a dataset of 17,131 posts. To estimate the proportion of researcher users in Group 1 and Group 2, we compared our dataset with the publicly available dataset published by Mongeon et al. ( 2022 , 2023 ). Using the author_id values from their 462,427 record dataset, we re-obtained usernames, identifying that 74,386 accounts had been deleted, leaving 388,041 active users. The verification process was conducted using the lowercase-converted usernames to ensure consistency across datasets. 3. Results 3.1 Score Distribution and Paper Classification Based on the aggregated scores of DIRECT X mentions and INDIRECT X mentions, we conducted descriptive statistical analyses, compared rankings between DIRECT X mentions and the combined scores (DIRECT + INDIRECT), and classified papers into four quadrants. The cumulative score of both mentions for the top 100 papers over the 1.5-year observation period reached 168,193, with DIRECT X mentions accounting for 58% and INDIRECT X mentions for 42%. This finding demonstrates that second-order citations exist on a comparable scale to first-order citations. Although 25% of the papers exhibited very low INDIRECT X mentions values (0–1), the maximum reached 8,625, exceeding that of DIRECT X mentions. In total, 19 papers recorded higher INDIRECT X mentions scores than their DIRECT X mentions counterparts (Table 1 ). The Spearman’s rank correlation coefficient between DIRECT and INDIRECT X mentions was 0.53, suggesting a moderate positive correlation but no strict proportional relationship between the two metrics. When comparing the combined score (DIRECT + INDIRECT X mentions) with the DIRECT X mentions, the minimum values remained almost unchanged — increasing only slightly from 252 to 254 — whereas the maximum value rose dramatically from 7,308 to 15,404, indicating a substantial expansion in the upper range (Table 1 ). Figure 1 presents the top 50 papers from the re-ranked list of all 100 papers based on their total combined scores. Among these 50 papers, 23 showed rank fluctuations of more than ten positions upward or downward. These substantial ranking shifts demonstrate that incorporating INDIRECT X mentions not only reveals the existence of second-order citations but fundamentally alters the evaluation of research impact, identifying influential papers that would otherwise be undervalued by conventional metrics relying solely on first-order citations. Table 1 Descriptive statistics of 100 papers Total Score in 1.5 years Min 25th Percentile 50th Percentile 75th Percentile Max Larger than the other DIRECT X mentions 97,626 252 318.75 499 933.75 7,308 81 INDIRECT X mentions 70,567 0 1 90 629.5 8,625 19 Total Score 168,193 254 379 726 1,856 15,404 - Figure 2 maps the 100 papers using INDIRECT X mentions on the x-axis and DIRECT X mentions on the y-axis, classifying them into four quadrants based on their mean values. Even though all target papers were among the top 100 in DIRECT X mentions, they were unevenly distributed across the four quadrants, with a majority (64 papers) concentrated in Q3, indicating that both DIRECT and INDIRECT X mentions exhibited a highly skewed distribution. This finding aligns with previous studies that have consistently reported highly skewed distributions in altmetric data (Banshal et al., 2022 ). Our results extend these findings by demonstrating that not only first-order citations (DIRECT X mentions) but also second-order citations (INDIRECT X mentions) exhibit similarly skewed characteristics. The remaining papers were relatively well distributed across the other quadrants. Papers in Q1 generally showed high values in both metrics, representing research with strong dissemination both within and beyond academic communities, whereas Q2 papers exhibited high DIRECT but low INDIRECT X mentions, reflecting primarily academic attention. Among these, 12 papers in Q4 were newly identified as influential works not sufficiently captured by the conventional metric based on DIRECT X mentions. The increases in their scores were particularly striking — as shown in Fig. 1 , ten of these twelve papers exhibited substantial upward rank shifts, rising by approximately 25 to 37 positions, with two papers newly entering the top ten. Table 2 shows the paper titles classified into quadrants Q1, Q2, and Q4. While topics do not clearly separate across quadrants, certain thematic tendencies can be observed. Q1 papers frequently featured keywords related to practical applications in high-stakes domains (e.g., "USMLE," "Medical Education," "Clinical Knowledge," "Finance"), evaluation of AI capabilities (e.g., "Performance," "Evaluating," "theory of mind," "Emergent Abilities"), and societal implications (e.g., "Labor Market Impact"). Q2 papers tended to emphasize technical methodologies (e.g., "deep reinforcement learning," "Multimodal Deep Learning"), domain-specific applications (e.g., "protein sequences," "Japanese Medical," "soccer skills"). Q4 papers exhibited more diverse characteristics, including technical documentation (e.g., "GPT-4 Technical Report"), workplace implications (e.g., "Generative AI at Work"), security and ethical concerns (e.g., "Indirect Prompt Injection," "Watermark," "Moral Self-Correction"), AI output evaluation (e.g., "abstracts generated by ChatGPT," "plagiarism detector"), and interdisciplinary applications (e.g., "ChemCrow," "technosignatures," "beat synchronization in rats"). While Q2 predominantly featured technical and domain-specific topics, the tendency for Q4 to cluster papers related to real-world applications of AI suggests that INDIRECT X mentions capture discussions centered not within academic communication but within broader public and media discourse. This pattern indicates that second-order citations reflect reactions from non-academic audiences and highlight the societal dimensions of research influence. Particularly noteworthy is the placement of the “GPT-4 Technical Report” in Q4. Despite being an OpenAI technical report outlining the foundational architecture of ChatGPT and having an exceptionally high citation count—2,043 citations as of November 2025—it was classified into Q4. While the papers with the highest citation counts, such as those published in PLOS Digital Health and Nature , appear in Q1, the fact that the “GPT-4 Technical Report” —a globally influential and highly cited work—fell into Q4 strongly underscores the necessity of measuring social dissemination through second-order citations. This finding serves as compelling evidence of the value and effectiveness of INDIRECT X mentions as an indicator for capturing societal-level engagement. 3.2 Time Series Analysis While the previous section examined the overall distribution and relative magnitude of DIRECT and INDIRECT X mentions, it remains unclear how these two indicators differ in their temporal diffusion patterns. Previous research has shown that Twitter mentions of scientific publications exhibit distinct temporal patterns (Priem & Costello, 2010 ; Fang et al., 2020). However, these findings have been based exclusively on first-order citations (DIRECT X mentions in our terminology). The temporal dynamics of second-order citations (INDIRECT X mentions) remain unexplored. Understanding whether second-order citations emerge and evolve differently from first-order citations is essential for capturing the process of knowledge diffusion both within and beyond academia, as well as for accurately measuring the dynamics of research impact. Therefore, this study conducted a time-series analysis of DIRECT X mentions and INDIRECT X mentions to elucidate their respective characteristics in terms of immediacy and long-term influence. Table 2 List of paper titles classified into quadrants Q1, Q2, and Q4 Q Paper Title (Journal / Collection) Citation Q1 1 2 3 5 7 8 9 11 12 14 16 “Sparks of Artificial General Intelligence: Early experiments with GPT-4” ( ArXiv ) “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models” ( ArXiv ) “Performance of ChatGPT on USMLE: Potential for AI-Assisted Medical Education Using Large Language Models” ( PLOS Digital Health ) “BloombergGPT: A Large Language Model for Finance” ( ArXiv ) “Large Language Models Encode Clinical Knowledge” ( Nature ) “Using deep learning to predict ideology from facial photographs: expressions, beauty, and extra-facial information” (Scientific Reports ) “Human-level play in the game of Diplomacy by combining language models with strategic reasoning” ( Science ) “The connectome of an insect brain” ( Science (New York, N.Y.) ) “Evaluating large language models in theory of mind tasks” ( Proceedings of the National Academy of Sciences of the United States of America ) “Are Emergent Abilities of Large Language Models a Mirage?” ( ArXiv ) “Dissociating language and thought in large language models” ( Trends in Cognitive Sciences ) 1,478 518 2,973 294 2,159 17 36 373 129 131 209 Q2 6 15 17 18 19 22 24 26 28 33 35 37 38 “Learning agile soccer skills for a bipedal robot with deep reinforcement learning” ( Science Robotics ) “A Survey of Large Language Models” ( ArXiv ) “Multimodal Deep Learning” ( International Conference on Machine Learning ) “Emergent autonomous scientific research capabilities of large language models” ( ArXiv ) “Superhuman artificial intelligence can improve human decision-making by increasing novelty” (Proceedings of the National Academy of Sciences of the United States of America ) “GPT detectors are biased against non-native English writers” (Patterns ) “Talking About Large Language Models” ( Communications of the ACM ) “Using AI to Implement Effective Teaching Strategies in Classrooms: Five Strategies, Including Prompts” ( SSRN Electronic Journal ) “Large language models generate functional protein sequences across diverse families” (Nature Biotechnology ) “Occupational Heterogeneity in Exposure to Generative AI” (SSRN Electronic Journal ) “Evaluating GPT-4 and ChatGPT on Japanese Medical Licensing Examinations” ( ArXiv ) “Assigning AI: Seven Approaches for Students, with Prompts” (SSRN Electronic Journal ) “Foundation models for generalist medical artificial intelligence” ( Nature ) 106 1,330 189 73 73 301 200 162 816 78 50 124 1,165 Q4 4 10 13 20 21 23 25 27 29 32 34 36 “GPT-4 Technical Report” ( ArXiv ) “Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection” ( Proceedings of the 16th ACM Workshop on Artificial Intelligence and Security ) “Generative AI at Work” ( The Quarterly Journal of Economics ) “A deep-learning search for technosignatures from 820 nearby stars” ( Nature Astronomy ) “ChemCrow: Augmenting large-language models with chemistry tools” (Nature Machine Intelligence ) “Comparing scientific abstracts generated by ChatGPT to original abstracts using an artificial intelligence output detector, plagiarism detector, and blinded human reviewers” ( BioRxiv ) “A Watermark for Large Language Models” ( International Conference on Machine Learning ) “The Capacity for Moral Self-Correction in Large Language Models” ( ArXiv ) “AI model GPT-3 (dis)informs us better than humans” ( Science Advances ) “Spontaneous beat synchronization in rats: Neural dynamics and motor entrainment” ( Science Advances ) “Is GPT-3 all you need for low-data discovery in chemistry?” ( ChemRxiv ) “The growing influence of industry in AI research” ( Science ) 2,043 190 673 34 336 387 110 44 170 28 26 131 Citation counts obtained via OpenAlex ( https://openalex.org/ ) as of November 13, 2025. 3.2.1 Verification of immediacy According to Priem and Costello ( 2010 ) 15% of first-order citations on X occur on the day of publication, and 40% within the first week. In this study, the reaction appeared even more rapidly: as shown in Table 3 , approximately 30% of DIRECT X mentions occurred on the publication day, and about 90% within the first week. In contrast, INDIRECT X mentions showed a slower response, with 15.8% appearing on the publication day and 39.5% within the first week. Consequently, the overall proportions of INDIRECT X mentions were close to the benchmark values reported by Priem and Costello ( 2010 ). Considering that traditional citation metrics typically require about a year to accumulate measurable counts, these results indicate that INDIRECT X mentions still exhibit considerable immediacy. Further analysis of the response timing for each paper revealed that in 77% of the cases, DIRECT X mentions appeared earlier than INDIRECT X mentions, while no papers exhibited earlier INDIRECT X mentions. The average difference in response time between the two indicators was 64.0 days (Table 3 ). Table 3 Elapsed days from publication to first post DIRECT X mentions INDIRECT X mentions *exclude 0 Proportion on the day of publication 30.0% 15.8% Proportion within one week after publication 90.0% 39.5% DIRECT X mentions is faster than INDIRECT 57 papers DIRECT X mentions is as same as INDIRECT 19 papers DIRECT X mentions is later than INDIRECT 0 papers Mean difference in elapsed days (INDIRECT - DIRECT) 64.0 days All items except “Proportion on the day of publication” and “Proportion within one week after publication” for DIRECT X mentions exclude papers with an INDIRECT X mentions value of zero, which applies to 24 out of the 100 papers. 3.2.2 Verification of long-term influence To determine whether first- and second-order citations differ in the persistence of their influence, we performed a longitudinal analysis of each paper. We calculated the number of days required for each paper to reach between 50% and 100% of its final score. Previous research by Fang et al. (2020) found that Twitter data accumulate very rapidly, with a half-life of only 13 days across multiple scientific fields, indicating that Twitter is among the fastest altmetric data sources in accumulating mentions shortly after publication. However, this finding was based on only first-order citations (DIRECT X mentions) Our results reveal a consistent pattern of delayed dissemination for INDIRECT X mentions across all thresholds (Table 4 ). At the 50% threshold, DIRECT X mentions required only 7.0 days (median) or 42.5 days (mean), whereas INDIRECT X mentions required substantially longer periods—41.5 days (median) and 103.9 days (mean). This gap widened progressively through the 60%, 70%, and 80% thresholds, demonstrating that second-order citations accumulated more slowly at each stage. By the results show that DIRECT X mentions reached 90% of their cumulative score after an average of 125.4 days (median = 66.5 days), whereas INDIRECT X mentions required a longer period, with an average of 188.4 days (median = 127 days) (Table 4 ). Across the 50–90% thresholds, INDIRECT X mentions consistently appeared later than DIRECT X mentions, reflecting the delayed and secondary nature of public dissemination through intermediary media. However, at the 100% threshold, DIRECT X mentions exhibited a longer tail, suggesting that academic users occasionally revisited or re-shared the papers long after their initial release. Table 4 The number of days required for each paper to reach between 50% and 100% of its final score DIRECT X mentions (days) INDIRECT X mentions (days) Median Mean Median Mean 50% 7.0 42.5 41.5 103.9 60% 10.0 46.4 52.0 113.2 70% 12.0 59.0 76.0 127.5 80% 31.0 78.7 89.0 152.3 90% 66.5 125.4 127.0 188.4 100% 502.5 460.3 378.5 297.9 Figure 3 The distribution of papers based on the difference in days to the 90% attainment (x-axis) and the difference in cumulative score between DIRECT and INDIRECT X mentions (y-axis). The numbers in each of the four quadrants, labeled A, B, C, and D, represent the number of papers distributed within each area, based on the axes where x = 0 and y = 0. Since the 24 papers with an INDIRECT X mentions value of zero are always classified into quadrant C, they are highlighted in orange and counted separately. Figure 3 illustrates the distribution of papers based on the difference in elapsed days (x-axis) and the difference in cumulative values (y-axis) between DIRECT and INDIRECT X mentions at the 90% threshold. The results reveal that in more than 85% of cases, INDIRECT X mentions required a longer time to converge, regardless of their absolute magnitude relative to DIRECT X mentions . These findings suggest that even after first-order citations plateau, second-order citations continue to contribute to the sustained social visibility of a paper, highlighting the crucial role of indirect information diffusion in reactivating or sustaining the public impact of research over time. 3.3 Network Analysis While the previous section revealed differences in the speed and persistence of dissemination between first- and second-order citations, it remains unclear who actually participates in these dissemination processes. Understanding whether the posts are made primarily by researchers or by broader non-academic audiences is crucial for evaluating research impact. In this chapter, we examine the possibility that second-order citations exert social influence within different audience spheres compared to first-order citations. Specifically, based on the actual post data of DIRECT X mentions and INDIRECT X mentions, we analyze user overlap and the proportion of researchers by constructing user networks consisting of posting users and mentioned users (i.e., users referenced through mentions, replies, or quotes). First, to assess the degree of network overlap between the two groups, the mean Jaccard index for each paper was calculated. The results revealed a remarkably low level of user overlap, with the average Jaccard index for posting users was 0.012, for mentioned users was 0.034, and for all users combined was 0.017. This indicates that the users who directly posted links to papers (first-order citations) and those who shared intermediary webpages (second-order citations) were almost entirely distinct. This tendency remained consistent across all four quadrants (Fig. 4 ). Next, we calculated the proportion of researcher users in both the first-order and second-order citation groups. Among posting users, Group 1 (DIRECT X mentions) comprised 9.9% researchers, whereas Group 2 (INDIRECT X mentions) comprised only 4.1%, approximately half the proportion of Group 1. This result indicates that first-order citations are primarily driven by researchers, while second-order citations are more strongly associated with non-academic communicators. Differences were also observed across quadrants. Quadrant 2, characterized by high DIRECT X mentions and low INDIRECT X mentions, exhibited the highest share of researchers at 10.8%, approximately twice that of Quadrant 4 (5.4%), which displayed the opposite pattern (low DIRECT and high INDIRECT X mentions). In Quadrant 4, however, the proportion of researchers within both Groups 1 and 2 reached the highest values—16.2% and 7.1%, respectively. Regarding mentioned users, although Quadrant 2 showed particularly high values, the overall proportions of researchers across both groups were generally comparable. Compared with posting users, Group 2 contained a relatively higher proportion of researcher accounts, yet there was almost no overlap between researcher users in Group 1 and Group 2 (Table 5 ). Figure 5 presents the top 20 nodes with the highest frequencies of occurrence among posting and mentioned users, summarizing the representative account classifications for both Group 1 and Group 2. The results indicate that the posting activity in Group 1 was influenced by the bot account “arxivabs”, which automatically replies with corresponding paper links. In addition, many of the top posting users in Group 1 were identified as researchers, most of whom were the authors of the referenced papers. In contrast, Group 2 included three researcher accounts; however, the majority consisted of general users and multiple news media sites, none of which were paper authors or media authors. Regarding the mentioned users, Group 1 was predominantly composed of researcher accounts, most of whom were again paper authors. The remaining top-mentioned users primarily consisted of research-related accounts, including research and development organizations (e.g., @openai, @penn ) and academic publication sites (e.g., @nature, @arxiv ). In contrast, the top-mentioned users in Group 2 were largely news outlets, with only one researcher account identified — a media article author. Several general users were also found to be authors of media articles, further illustrating the distinctive composition of Group 2’s dissemination network. Table 5 Proportion of researcher among all users Total Users Researcher (overlapping researcher) Percentage of researchers 1 Positing Users Group1 2,512 249 9.9% Group2 9,349 386 4.1% Q1 Group1 Group2 5,055 1,336 3,791 203 (5) 101 107 4.0% 7.6% 2.8% Q2 Group1 Group2 713 333 395 77 (5) 54 28 10.8% 16.2% 7.1% Q3 Group1 Group2 845 265 593 48 (3) 34 17 5.7% 12.8% 2.9% Q4 Group1 Group2 6,470 793 5,800 348 (13) 83 278 5.4% 10.5% 4.8% 2 Referenced Users Group1 1,855 176 9.5% Group2 1,694 141 8.3% Q1 Group1 Group2 1,605 1,051 600 113 (13) 80 46 7.0% 7.6% 7.7% Q2 Group1 Group2 262 168 101 47 (2) 42 7 17.9% 25.0% 6.9% Q3 Group1 Group2 250 119 141 25 (3) 17 11 10.0% 14.3% 7.8% Q4 Group1 Group2 1,611 572 1,094 146 (10) 55 101 9.1% 9.6% 9.2% The numbers in parentheses for Q1–Q4 indicate the counts of overlapping researcher users between Group1 and Group2. 4. Discussion This study proposed INDIRECT X mentions, a novel indicator based on second-order citations, to complement traditional DIRECT X mentions, which capture only first-order dissemination of academic papers. The analysis demonstrated that, consistent with previous research, second-order citations are also widespread in the AI domain. Despite selecting the top 100 papers based on DIRECT X mentions, the volume of INDIRECT X mentions was found to be of comparable magnitude. When combining both indicators into a total score, the overall distribution expanded significantly, allowing papers that achieved high levels of societal attention to be more clearly distinguished. The time-series analysis revealed that the two indicators exhibit distinct diffusion characteristics. Specifically, INDIRECT X mentions showed slower initial reactions but a greater tendency for long-term dissemination over time. These differences likely stem from the structural mechanism of data aggregation. Since INDIRECT X mentions can only be counted once intermediary webpages—such as news articles or blogs—have been created, they tend to appear later but may continue to spread as long as new intermediary webpages emerge. In contrast, DIRECT X mentions reflected an immediate surge in attention, particularly at key events such as the release of a paper’s second edition, where a “reactivation” phenomenon was observed (Fig. 6 ). These findings suggest that indirect dissemination via intermediary webpages functions as a mechanism for sustaining public engagement with research beyond the initial citation period. Consequently, INDIRECT X mentions can be interpreted as an indicator that captures sustained societal engagement, a dimension previously overlooked by conventional altmetric measures. The network analysis further revealed that first- and second-order citations are generated by largely distinct groups of users. Compared to first-order citations, second-order citations involved substantially fewer researchers. This pattern mirrors the findings of Alperin et al. ( 2024 ) in their COVID-19 case study, confirming that such a distinction is a consistent feature of second-order dissemination. Examination of the top posting and mentioned users indicated that DIRECT X mentions were primarily generated by bots (e.g., arxivabs ) and paper authors, whereas INDIRECT X mentions were dominated by general users, journalists, and news media accounts with limited direct involvement in research. The difference in mentioned-user composition likely reflects posting conventions on X, where users often credit the original source—either a paper or a media article—by tagging its hosting site or author account. Given that users on X form distinct follower networks, variations in posting-user composition lead to markedly different diffusion pathways. Thus, both DIRECT and INDIRECT X mentions, which incorporate repost and quote activity, represent audience-specific patterns of engagement with research outputs. This interpretation is further supported by the quadrant-based analysis. Papers in Quadrant 2 (Q2), characterized by high DIRECT and low INDIRECT X mentions, had the highest proportion of researchers (10.8%), whereas Quadrant 4 (Q4), showing the opposite pattern, exhibited roughly half that rate (5.4%). Q2 primarily comprised technical studies such as “Learning Agile Soccer Skills for a Bipedal Robot with Deep Reinforcement Learning” and “Multimodal Deep Learning” , while Q4 included socially and ethically oriented papers such as “GPT-4 Technical Report” , “Generative AI at Work” , and “The Capacity for Moral Self-Correction in Large Language Models.” These results suggest that INDIRECT X mentions more effectively capture the societal dimension of research impact that traditional altmetrics often overlook. The fact that the “GPT-4 Technical Report,” which had been cited 2,043 times as of November 2025, was classified into Q4—despite its exceptionally high scholarly influence and relatively low DIRECT X mentions—strongly reinforces the necessity of detecting the societal impact generated through second-order citations. This finding serves as a clear demonstration of the effectiveness of INDIRECT X mentions. Moreover, it suggests that research fields attracting broad public attention may exhibit distinctive dissemination patterns that differ from those observed in more traditional academic domains. Papers highly ranked on both indicators —such as “Sparks of Artificial General Intelligence” , “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models” , and “Performance of ChatGPT on USMLE: Potential for AI-Assisted Medical Education Using Large Language Models” —represent research that successfully engaged both academic and public audiences. Nevertheless, this study is limited by its focus on the top 100 papers in the AI domain. Further work is required to assess cross-disciplinary reproducibility and to establish a generalized classification framework that integrates both indicators. Future research should also analyze the content of intermediary webpages to better understand how academic findings are reframed and disseminated in social and media contexts, thereby enabling a more comprehensive evaluation of the societal influence of research. 5. Conclusion This study conceptualized and operationalized second-order citations as INDIRECT X mentions, and conducted a comparative validation against the existing first-order citation–based indicator, DIRECT X mentions, which aggregates the number and dissemination counts of direct posts linking to academic papers. Our findings reveal that in the AI domain, a substantial amount of indirect dissemination occurs through news articles and blogs, involving user groups that more closely represent general audiences. Furthermore, second-order citations exhibited a tendency to sustain engagement on X over a longer period, thereby providing clearer evidence of their importance for accurately capturing the dynamics of research dissemination on social media. While this study is limited to a case analysis of 100 AI-related papers, our initial hypothesis—that understanding the true dissemination of academic outputs requires accounting not only for direct but also indirect citations—was supported within this scope. In rapidly evolving and socially visible research fields such as AI, where both academic and non-academic interest is particularly high, the altmetrics initiative of “visualizing the real-time societal impact of research” holds significant value. In this respect, this study provides meaningful insights by validating the visualization and applicability of both indicators in the AI field. Although further empirical validation is needed for the proposed INDIRECT X mentions, this study contributes by establishing its definition as a measurable indicator and by presenting a reproducible and transparent aggregation framework based on the existing Altmetrics’s Mentions data. The proposed procedure can be replicated using publicly accessible data categories (News and Blogs) and standard APIs, ensuring methodological consistency and facilitating future comparative studies. Previous research has shown that intermediary webpages provided by data suppliers exhibit language-based bias (Ortega, 2019b). A similar pattern was observed in the present dataset: both indicators were strongly skewed toward English, and particularly for Japanese, a substantial discrepancy was evident between the proportions of DIRECT X mentions and INDIRECT X mentions. Furthermore, 18 out of the 100 papers had zero intermediary webpages, suggesting the need for careful validation of the dataset’s reliability. Given that the overlap of intermediary webpage coverage across providers is known to be low, the use of multiple data providers is recommended (Haustein et al., 2013 ). The two indicators proposed in this study—DIRECT X mentions and INDIRECT X mentions—hold considerable potential for future applications. Although further investigation is required to establish standardized methods for paper classification, the differences in the intrinsic characteristics of these indicators suggest promising opportunities for developing classification frameworks that leverage both metrics. Existing research has extensively examined the relationship between DIRECT X mentions and conventional indicators of academic success, such as citation counts, yet no clear or consistent correlation has been confirmed (Haustein et al., 2013 ; Bornmann, 2015 ; Mazurek et al., 2022 ; Howoldt et al., 2023 ). In contrast, INDIRECT X mentions may represent a metric that captures a distinct dimension of societal influence not reflected in traditional indicators, thereby offering substantial potential as a predictive measure. Both indicators demonstrated convergence earlier than 1.5 years after publication—a notably shorter timescale than that of conventional citations, which typically begin accumulating after one year or more. Furthermore, the case of the “GPT-4 Technical Report” in the Q4 classification revealed the presence of large-scale public discourse occurring on X that was not captured by first-order–based indicators. This finding underscores the significant role of INDIRECT X mentions in reflecting public engagement within the broader context of citizen participation in science and highlights important directions for future research. Declarations Author Contribution Ai Kishimoto wrote the main manuscript text and prepared all Figures and Tables. All authors reviewed the manuscript. Data Availability All data supporting the findings of this study are available from National Graduate Institute for Policy Studies. References Alhoori, H., & Furuta, R. (2014). Do Altmetrics Follow the Crowd or Does the Crowd Follow Altmetrics? IEEE Press , 375–378. Alperin, J. P., Fleerackers, A., Riedlinger, M., & Haustein, S. (2024). Second-order citations in altmetrics: A case study analyzing the audiences of COVID-19 research in the news and on social media. Quantitative Science Studies , 5(2), 366–382. https://doi.org/10.1162/qss_a_00298 Banshal, S. K., Gupta, S., Lathabai, H. H., & Singh, V. K. (2022). Power laws in altmetrics: An empirical analysis. Journal of Informetrics , 16 , 101309. https://doi.org/10.1016/j.joi.2022.101309 Bornmann, L. (2015). Alternative metrics in scientometrics: A meta-analysis of research into three altmetrics. Scientometrics , 103 (3), 1123–1144. Bornmann, L., & Haunschild, R. (2019). Societal impact measurement of research papers. In W. Glänzel, H. F. Moed, U. Schmoch, & M. Thelwall (Eds.), S pringer handbook of science and technology indicators (pp. 609–632). Springer Nature. Brody, T., Harnad, S., & Carr, L. (2006). Earlier Web usage statistics as predictors of later citation impact. Journal of the American Society for Information Science and Technology , 57(8), 1060–1072. https://doi.org/10.1002/asi.20373 Costas, R., Mongeon, P., Ferreira, M. R., van Honk, J., & Franssen, T. (2020). Large-scale identification and characterization of scholars on Twitter. Quantitative Science Studies , 1(2), 771–791. https://doi.org/10.1162/qss_a_00047 Fang, Z., & Costas, R. (2020). Studying the accumulation velocity of altmetric data tracked by Altmetric.com. Scientometrics , 123 , 1077–1101. https://doi.org/10.1007/s11192-020-03405-9 Gingras, Y. (2014). Criteria for evaluating indicators. In B. Cronin & C. R. Sugimoto (Eds.), Beyond bibliometrics: Harnessing multidimensional indicators of scholarly impact (pp. 109–125). MIT Press. Haustein, S. (2019). Scholarly twitter metrics. In Springer handbook of science and technology indicators (pp. 729–760). Cham: Springer International Publishing. Haustein, S., Peters, I., Sugimoto, C. R., Thelwall, M., & Larivière, V. (2013) Tweeting biomedicine: an analysis of tweets and citations in the biomedical literature. Journal of the Association for Information Science and Technology , 65 (4), 656–669. https://doi.org/10.1002/asi.23101 Howoldt, D., Kroll, H., Neuhäusler, P., & Feidenheimer, A. (2023). Understanding researchers’ Twitter uptake, activity and popularity—an analysis of applied research in Germany. Scientometrics , 128 (1), 325–344. Ke, Q., Ahn, Y., & Sugimoto, C. R.(2017). A systematic identification and analysis of scientists on Twitter. PLoS ONE 12(4): e0175368. https://doi.org/10.1371/journal.pone.0175368 Mazurek, G., Gorska, A., Korzynski, P., & Silva, S. (2022). Social networking sites and researcher’s success. Journal of Computer Information Systems , 62 (2), 259–266. Mongeon, P., Bowman, T., & Costas, R. (2022). Open dataset of scholars on Twitter. Zenodo . https://doi.org/10.5281/zenodo.7013518 Mongeon, P., Bowman, T. D., & Costas, R. (2023). An open data set of scholars on Twitter. Quantitative Science Studies , 4(2), 314–324. https://doi.org/10.1162/qss_a_00250 Ortega,J. L. (2018). Reliability and accuracy of altmetric providers: a comparison among Altmetric.com, PlumX and Crossref Event Data. Ortega, J. L. (2020a). Altmetrics data providers: A meta-analysis review of the coverage of metrics and publication. El Profesional de La Información , 29(1). https://doi.org/10.3145/epi.2020.ene.07 Ortega,J. L. (2020b). Blogs and news sources coverage in altmetrics data providers: a comparative analysis by country, language, and subject. Scientometrics , 122, 555–572. Springer Nature. Priem, J., & Costello, K. L. (2010). How and why scholars cite on Twitter. Proceedings of the American Society for Information Science and Technology, 47 (1), 1–4. https://doi.org/10.1002/meet.14504701201 Priem, J., Taraborelli,D., Groth, D., & Neylon,C.(2010). Altmetrics: A manifesto. Priem, J. (2014). Altmetrics. In B. Cronin & C. R. Sugimoto (Eds.), Beyond bibliometrics: Harnessing multidimensional indicators of scholarly impact (pp. 263–288). MIT Press. Yu, H., Xiao, T., Xu, S., & Wang, Y. (2019). Who posts scientific tweets? An investigation into the productivity, locations, and identities of scientific tweeters. Journal of Informetrics, 13(3), 841–855. van Raan, A. F. J. (2019). Measuring science: Basic principles and application of advanced bibliometrics. In W. Glänzel, H. F. Moed, U. Schmoch, & M. Thelwall (Eds.), Springer handbook of science and technology indicators (pp. 237–280). Springer Nature. Wouters, P. (2014). The citation: From culture to infrastructure. In B. Cronin & C. R. Sugimoto (Eds.), Beyond bibliometrics: Harnessing multidimensional indicators of scholarly impact (pp. 47–66). MIT Press. Wouters, P., Zahedi, Z., & Costas, R. (2019). Social media metrics for new research evaluation.In W. Glänzel, H. F. Moed, U. Schmoch, & M. Thelwall (Eds.), Springer handbook of science and technology indicators (pp. 687–713). Springer Nature. Zahedi, Z., Costas, R. (2018). General discussion of data quality challenges in social media metrics: Extensive comparison of four major altmetric data aggregators. PloS one , 13 (5), e0197326. https://doi.org/10.1371/journal.pone.0197326 Zhang, L., Gou, Z., Fang, Z., Sivertsen, G., & Huang, Y. (2023). Who tweets scientific publications? A large-scale study of tweeting audiences in all areas of research. Journal of the Association for Information Science and Technology, 74, 1485–1497. https://doi.org/10.1002/asi.24830 Footnotes https://www.altmetric.com/ https://www.elsevier.com/insights/metrics/plumx https://profiles.impactstory.org/ Altmetric.com, https://www.altmetric.com/blog/200-million-mentions-altmetric-has-now-officially-tracked-over-200-million-mentions-of-nearly-20-million-research-outputs/ Altmetric.com - X (formerly Twitter) https://help.altmetric.com/support/solutions/articles/6000235926-x-formerly-twitter- https://chatgpt.com/ https://nodexl.com/ Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8121384","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":561829245,"identity":"fa11c284-c52e-48d0-acc2-92bcc362d790","order_by":0,"name":"Ai Kishimoto","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABA0lEQVRIie3PsUoDMRjA8YRApkjW3nK+QsqtvkmXZKlT9gxtiQh3i9C1gtRXqMvN3xFIl0DXQgfdXAsuHQTN6SAOd6ebYP5DCCE/kg+hVOpPRhEgccE4IeBeTTwgxP6AmGmeVVQBCy3BwwSh4AqxYwWcle3JAOFVqQCXRN1eUwnZej7hVSQnU3eSUfAQCVXLOAuM661eOWzxTTh0P7O/tJGwz1dU7bWNhOCym5w/PrdkpDaOCWjuvL4fImJPAVAQhYikubIzvRki4zCVII3Ms/gxhzzoh0iavlnyrS+OR/HGOHfuBc0Wer1zzdPJ9IzfJr+27mOF/vvfW/zmciqVSv2T3gEV1GWQZ9XlEQAAAABJRU5ErkJggg==","orcid":"","institution":"National Graduate Institute for Policy Studies","correspondingAuthor":true,"prefix":"","firstName":"Ai","middleName":"","lastName":"Kishimoto","suffix":""},{"id":561829247,"identity":"c76e7613-fdb5-4e5b-8924-6d13995244f8","order_by":1,"name":"Takayuki Hayashi","email":"","orcid":"","institution":"National Graduate Institute for Policy Studies","correspondingAuthor":false,"prefix":"","firstName":"Takayuki","middleName":"","lastName":"Hayashi","suffix":""}],"badges":[],"createdAt":"2025-11-15 10:38:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8121384/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8121384/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98626252,"identity":"e3deb794-27a1-41d7-abc2-2fe1e007cee0","added_by":"auto","created_at":"2025-12-19 17:09:39","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1236953,"visible":true,"origin":"","legend":"","description":"","filename":"ProposalofINDIRECTXMentionsasanAltmetricsIndicatorDisseminationofResearchPapersonXviaWebNewsandBlogs.docx","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/558008b488c742164719685e.docx"},{"id":98545074,"identity":"f82783e2-32c0-4ddb-8ae4-087aee3bec02","added_by":"auto","created_at":"2025-12-18 18:41:33","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4158,"visible":true,"origin":"","legend":"","description":"","filename":"5d753c66e1af4312959c2150bd308007.json","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/8a9f304761ea82f26153bcdb.json"},{"id":98545078,"identity":"5a11b09f-4992-4b55-91b0-89b7eecd6bc6","added_by":"auto","created_at":"2025-12-18 18:41:34","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":119329,"visible":true,"origin":"","legend":"","description":"","filename":"5d753c66e1af4312959c2150bd3080071enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/df4c3c44269d4bb2552d1c51.xml"},{"id":98625696,"identity":"b51445dc-4cf4-498c-a968-88ecd8fd77cb","added_by":"auto","created_at":"2025-12-19 17:09:16","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":24899,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/066b69238cb5d1dce7eb5fea.png"},{"id":98545086,"identity":"cedf9c6c-526e-4140-a5f1-aedb7599fe44","added_by":"auto","created_at":"2025-12-18 18:41:34","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13896,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/5c0aab1987f5f1aab6b54d1b.png"},{"id":98545081,"identity":"c516f1f5-7a47-42d5-aa6c-e3f912a3dbbe","added_by":"auto","created_at":"2025-12-18 18:41:34","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31886,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/614871943b2cc2aa6c378be1.png"},{"id":98545088,"identity":"5b83fc8d-b5b0-4648-92e5-ed376f0a0b79","added_by":"auto","created_at":"2025-12-18 18:41:34","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":115641,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/61f3d459dfc91b1e6583dc61.png"},{"id":98545085,"identity":"bd5b760a-bee3-440e-a296-e26db38b1c8b","added_by":"auto","created_at":"2025-12-18 18:41:34","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":80507,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/4aacf4c5891c06429ab4dbb2.png"},{"id":98626002,"identity":"85838cab-b5ad-43a3-af61-754af739fe97","added_by":"auto","created_at":"2025-12-19 17:09:28","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26989,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/51ead297e82c8c6407249b45.png"},{"id":98545090,"identity":"14a989bd-0f8f-40af-b334-d161ea1f5fb8","added_by":"auto","created_at":"2025-12-18 18:41:34","extension":"xml","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":119209,"visible":true,"origin":"","legend":"","description":"","filename":"5d753c66e1af4312959c2150bd3080071structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/deb8c442782a41b8fa13c3f1.xml"},{"id":98626050,"identity":"f3e07b00-6dd2-41f5-96a6-7194e73aed2e","added_by":"auto","created_at":"2025-12-19 17:09:29","extension":"html","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":132684,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/69b42f83dd6c3a6c2f98e088.html"},{"id":98545076,"identity":"e7f44f7a-15e9-4d8b-9b73-24b093d9e972","added_by":"auto","created_at":"2025-12-18 18:41:33","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":98924,"visible":true,"origin":"","legend":"\u003cp\u003eTOP50 Papers from the re-ranked list of all 100 papers based on their total combined scores. Parentheses indicate the rankings under the conventional metric. Only papers with rank differences of ±10 or more compared to the original ranking are labeled above the bars. The colors below each bar indicate the corresponding quadrants classified in Fig. 2.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/caa9bc583741864af3058dd1.png"},{"id":98626942,"identity":"dc8da810-df7b-4ba5-9cb7-94a82e073ded","added_by":"auto","created_at":"2025-12-19 17:10:03","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":48450,"visible":true,"origin":"","legend":"\u003cp\u003eScatter plot of the 100 papers. The x-axis represents INDIRECT X mentions and the y-axis represents DIRECT X mentions. The papers are classified into four quadrants based on the mean values of both scores.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/5a486855aef420e5cead33bc.png"},{"id":98626149,"identity":"9830fa21-5e53-4721-99b4-75363d3216b5","added_by":"auto","created_at":"2025-12-19 17:09:34","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":161367,"visible":true,"origin":"","legend":"\u003cp\u003eThe distribution of papers based on the difference in days to the 90% attainment (x-axis) and the difference in cumulative score between \u003cem\u003eDIRECT\u003c/em\u003e and \u003cem\u003eINDIRECT X mentions\u003c/em\u003e (y-axis). The numbers in each of the four quadrants, labeled A, B, C, and D, represent the number of papers distributed within each area, based on the axes where x = 0 and y = 0. Since the 24 papers with an INDIRECT X mentions value of zero are always classified into quadrant C, they are highlighted in orange and counted separately.\u003c/p\u003e","description":"","filename":"floatimage325.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/f736b0733cd187c8e65351e1.png"},{"id":98545080,"identity":"e743d890-91e5-45cc-86f4-da847a58cc0d","added_by":"auto","created_at":"2025-12-18 18:41:34","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":364811,"visible":true,"origin":"","legend":"\u003cp\u003eThe mean Jaccard index values for each paper were calculated separately for posting users, mentioned users, and all users combined. The mean Jaccard index values for each of the four quadrants are also shown.\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/9631c93559b5de82afc57ad3.jpeg"},{"id":98626166,"identity":"d49ce50e-f825-4eb9-a9b8-78764f0fe9b3","added_by":"auto","created_at":"2025-12-19 17:09:35","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":424485,"visible":true,"origin":"","legend":"\u003cp\u003eThe top 20 posting and referenced user nodes and account classifications for Groups 1 and 2. Researcher accounts were identified primarily using the dataset by Mongeon et al. (2022), supplemented with additional accounts whose bio descriptions explicitly indicated academic affiliations or doctoral credentials (PhD)\u003c/p\u003e","description":"","filename":"floatimage55.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/197143ce63fef38843f68289.png"},{"id":98625467,"identity":"e85f781d-bbb0-4cdd-bcf5-86d51fe6cbca","added_by":"auto","created_at":"2025-12-19 17:09:08","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":197988,"visible":true,"origin":"","legend":"\u003cp\u003eCumulative data of DIRECT X mentions over 1.5 years since the publication dates of 100 papers. The x-axis represents the number of elapsed days since publication, and the y-axis represents the cumulative DIRECT X mentions. Points showing sharp score increases on the dates of second or later editions of the papers are highlighted with red circles.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/3c3909b07ff7d1b32dfa82e4.png"},{"id":98775343,"identity":"d82bacdf-53bc-401c-8944-8ffa48716cee","added_by":"auto","created_at":"2025-12-22 12:19:29","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2138457,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8121384/v1/e16f306f-bc3c-425e-b1c7-ea3e6488f4b5.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Proposal of INDIRECT X Mentions as an Altmetrics Indicator: Dissemination of Research Papers on X via Web News and Blogs","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe evaluation of academic research has traditionally been based on citation counts and citation-based metrics. These indicators have been widely used not only in academia but also in policymaking. However, concerns have been raised regarding their reliability (Gingras, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; van Raan, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Wouters, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), time lag before citation counts accumulate (Priem, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Brody \u0026amp; Harnad, 2006) and inability to measure the societal impact of research beyond the academic community (Bornmann \u0026amp; Haunschild, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Priem, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Since the 2000s, the widespread use of the Internet has led to an increase in the need for new evaluation metrics, and Priem et al. (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) proposed the concept of \"altmetrics\". Currently, the practical application of altmetric indicators has been achieved by multiple data providers, most notably Altmetric.com\u003csup\u003e3\u003c/sup\u003e, PlumX\u003csup\u003e4\u003c/sup\u003e and Impactstory\u003csup\u003e5\u003c/sup\u003e. As of 2022, Altmetric.com had successfully tracked more than 200\u0026nbsp;million mentions associated with approximately 20\u0026nbsp;million research outputs, thereby demonstrating its capability to visualize the societal impact of scholarly papers in real time that cannot be measured solely by citation counts\u003csup\u003e6\u003c/sup\u003e (Condon,2022).\u003c/p\u003e \u003cp\u003eAmong altmetrics data sources, X (formerly Twitter) has drawn particular attention because of its data scale (Alhoori \u0026amp; Furuta, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) and potential as an indicator for measuring the societal impact of research (Wonters et al., 2019). Notably, academic discussions on X have been increasing, and further enhancing its value as a data source (Yu et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Priem and Costello (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) stated that the dissemination of academic papers on X involves first-order citations, which hyperlink directly to academic resources, and second-order citations, which hyperlink to intermediary webpages such as news articles or blogs that describe and link to the resources. Currently, both the X-based indicators provided by data providers\u003csup\u003e7\u003c/sup\u003e and many previous studies tend to focus solely on first-order citations. However, it has been shown that users who post first-order citations differ from the general population, exhibiting a bias toward individuals with academic backgrounds (Haustein, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Costas, 2020; Zhang, 2023). Conversely, several prior studies have suggested that second-order citations may exist on a comparable or even larger scale than first-order citations (Priem \u0026amp; Costello, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2010\u003c/span\u003e, Ke, at al., 2017). A recent case study by Alperin et al. (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) examined COVID-19 research by focusing on five prominent English-language news outlets\u0026mdash; BBC, MSN, The New York Times, The Guardian, and The Washington Post. They analyzed the second-order citations generated by news articles published in these outlets and found that such citations involved approximately twice as many unique posting users and engagement counts as first-order citations. Moreover, the user groups associated with first- and second-order citations showed minimal overlap, with the latter group containing a markedly lower proportion of researchers. These findings illustrate how scientific knowledge was disseminated to the broader public in the context of COVID-19, where public interest in science was particularly high. However, in more general research domains, knowledge is likely to be disseminated not only through major news outlets but also via a wider range of online media such as specialized news sites and blogs. Therefore, it is essential to evidence from diverse research domains to clarify how second-order citations differ from first-order citations in terms of frequency, user groups, and diffusion dynamics.\u003c/p\u003e \u003cp\u003eIn this study, we focus on second-order citations and analyze them using a newly defined indicator, which we refer to as INDIRECT X mentions, that encompasses a broader range of online media including diverse news sites and blogs. To clearly distinguish between the two indicators, we define DIRECT X mentions as a conventional metric that captures the direct dissemination of research papers through academic paper websites via first-order citations, and INDIRECT X mentions as a new metric that captures indirect dissemination through intermediary webpages via second-order citations.\u003c/p\u003e \u003cp\u003eAs a case study, this research focuses on the field of artificial intelligence (AI). In June 2018, OpenAI introduced GPT, and on November 30, 2022, the organization released ChatGPT\u003csup\u003e8\u003c/sup\u003e\u0026mdash; a web-based conversational service powered by GPT-3.5\u0026mdash; to the general public. The system rapidly gained unprecedented attention, surpassing 100\u0026nbsp;million users within just two months of its release. This breakthrough marked a transformative moment that propelled artificial intelligence (AI) into a globally significant research domain attracting broad interest not only from the academic community but also from industry and the general public. Despite this surge in societal attention, the social impact of AI-related research on X has not been sufficiently examined. To address this gap, the present study focuses on academic papers published between November 2022\u0026mdash;the month ChatGPT was released\u0026mdash;and April 2023. From this set, we selected 100 AI-related papers with the highest DIRECT X mentions and conducted a comparative analysis of two indicators\u0026mdash;DIRECT and INDIRECT X mentions\u0026mdash;over a 1.5-year period following publication. Although these papers are typically regarded as highly influential according to conventional altmetric rankings, this study demonstrates that incorporating INDIRECT X mentions can lead to a substantially different evaluation of their impact. By addressing the following research questions, we aim to clarify the existence and distinct diffusion patterns of paper dissemination via second-order citations that are not captured by first-order citations, emphasize the importance of measuring such second-order dissemination on X, and contribute to improving the validity and comprehensiveness of altmetric indicators.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eRQ1. What kinds of dissemination patterns can be identified when INDIRECT X mentions are incorporated into existing metrics (Score Aggregation Analysis)?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eRQ2. How do the temporal dynamics of paper dissemination differ between first-order and second-order citations (Time-Series Analysis)?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eRQ3. How does the user network differ between DIRECT and INDIRECT X mentions (Network Analysis)?\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e"},{"header":"2. Method and data","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Method for Collecting INDIRECT X mentions\u003c/h2\u003e \u003cp\u003eThe existing indicator, DIRECT X mentions, is defined as the sum of the total number of posts that directly hyperlink to the webpage of an academic paper (first-order citations) and their dissemination counts, including reposts and quotes. To enable comparative analysis, the new indicator proposed in this study, INDIRECT X mentions, is constructed in a manner similar to the existing metric. It sums the total number of posts that contain hyperlinks to intermediary webpages referring to academic papers (second-order citations) and their corresponding dissemination counts, including reposts and quotes. The aggregation of second-order citations was conducted using a two-step process. First, intermediary webpage data were collected using the \u0026ldquo;Mentions\u0026rdquo; category within the Altmetrics database, which aggregates News and Blogs data. Second, posts on X that included hyperlinks to these intermediary webpages were retrieved via the X API, and their reposts and quotes were counted. By aggregating these data, the INDIRECT X mentions score was systematically quantified. Through this method, the study establishes a consistent and comprehensive framework for measuring INDIRECT X mentions, enabling the systematic quantification of second-order citation-based dissemination.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Data on Papers, DIRECT X mentions, and INDIRECT X mentions\u003c/h2\u003e \u003cp\u003eIn this study, we used data provided by Altmetric.com to select the target papers and to obtain both the DIRECT X mentions scores and the intermediary webpage data. Although several data providers supply altmetric data, their data collection methodologies and coverage vary considerably. Among them, Altmetric.com is known for its comprehensive coverage of X (formerly Twitter) data and intermediary webpage data (Ortega, J. L., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Ortega, J. L., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2020a\u003c/span\u003e; Zahedi et al., 2018). Since the X score provided by Altmetric.com corresponds to the DIRECT X mentions defined in this study\u0026mdash;which include only first-order citations\u0026mdash;we used Altmetric.com\u0026rsquo;s existing X metrics as the basis for paper selection.\u003c/p\u003e \u003cp\u003eThis study focuses on the field of artificial intelligence (AI). Unlike COVID-19 by Alperin et al. (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), which represented an exceptional crisis situation with extraordinarily high public urgency, AI represents a more typical yet increasingly prominent research domain that attracts substantial interest from diverse audiences including the general public, technology practitioners, policymakers, and researchers. This makes AI particularly suitable for examining the dissemination patterns of scientific knowledge under more generalizable conditions. Target papers in the AI field were identified using the Altmetric.com subject codes (4602, 4611) and a keyword search applied to titles, including \u003cem\u003eAI, Artificial Intelligence, Deep Learning, GPT, LLM\u003c/em\u003e, and \u003cem\u003eLarge Language Model\u003c/em\u003e. The publication period was restricted to papers published between November 1, 2022, and April 30, 2023. For papers available on arXiv, the date of appearance on arXiv was treated as the official publication date because these preprints are often publicly disseminated and cited well before formal journal publication, thereby representing the earliest point of online visibility.\u003c/p\u003e \u003cp\u003eFrom the collected results, the top 100 papers were selected based on the DIRECT X mentions values measured 1.5 years after publication. Subsequently, for these 100 papers, we obtained daily time-series data of DIRECT X mentions and a list of intermediary webpages (URLs of Web News and Blogs) from Altmetric.com. As a result, 4,473 intermediary webpages were collected. For second-order citation data, we used NodeXL\u003csup\u003e9\u003c/sup\u003e, a social network analysis tool for collecting and visualizing social media data, to collect posts on X that contained URLs linking to the intermediary webpages. These collected posts numbered 28,926 in total. Based on these data, we aggregated the total number of posts, reposts, and quotes to calculate the INDIRECT X mentions scores for the 100 target papers.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Data for Network Analysis\u003c/h2\u003e \u003cp\u003eThe dataset for the network analysis included only posts containing relevant links and excluded its reshare posts. Each post consisted of a posting user and one or more mentioned users, forming a user interaction network. To clarify communication patterns between users, posts in which the posting user and the mentioned user were identical (such as self-replies or self-quotes) were excluded from the analysis. Usernames were converted to lowercase strings to ensure consistency and to check for duplicates. For analytical purposes, the network formed by DIRECT X mentions was defined as Group 1, and that formed by INDIRECT X mentions was defined as Group 2.\u003c/p\u003e \u003cp\u003eDue to data access limitations and significant cost constraints of the X API\u0026mdash;particularly for retrieving posts linking to intermediary webpages and their associated metadata\u0026mdash;a subset of papers was selected for this analysis. To better capture the characteristics of papers classified into the four quadrants in the RQ1 analysis, In total, 17 papers were included as the target dataset. While the resulting sample is relatively small, it was strategically selected to ensure diversity in dissemination patterns, as explained later.\u003c/p\u003e \u003cp\u003eFor Group 1 (first-order citations), data were recollected using the X API based on the tweet_id values provided by Altmetric.com. From the aggregated data, 20,328 tweet_ids were identified. However, due to privacy settings and account deletions, detailed post data could be retrieved for 20,107 posts via the X API. After filtering for posts within the observation period, 5,090 posts remained as the final dataset for analysis. For Group 2 (second-order citations), data were extracted from the collection used in the INDIRECT X mentions aggregation, restricted to the selected papers, resulting in a dataset of 17,131 posts.\u003c/p\u003e \u003cp\u003eTo estimate the proportion of researcher users in Group 1 and Group 2, we compared our dataset with the publicly available dataset published by Mongeon et al. (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Using the author_id values from their 462,427 record dataset, we re-obtained usernames, identifying that 74,386 accounts had been deleted, leaving 388,041 active users. The verification process was conducted using the lowercase-converted usernames to ensure consistency across datasets.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Score Distribution and Paper Classification\u003c/h2\u003e \u003cp\u003eBased on the aggregated scores of DIRECT X mentions and INDIRECT X mentions, we conducted descriptive statistical analyses, compared rankings between DIRECT X mentions and the combined scores (DIRECT\u0026thinsp;+\u0026thinsp;INDIRECT), and classified papers into four quadrants. The cumulative score of both mentions for the top 100 papers over the 1.5-year observation period reached 168,193, with DIRECT X mentions accounting for 58% and INDIRECT X mentions for 42%. This finding demonstrates that second-order citations exist on a comparable scale to first-order citations. Although 25% of the papers exhibited very low INDIRECT X mentions values (0\u0026ndash;1), the maximum reached 8,625, exceeding that of DIRECT X mentions. In total, 19 papers recorded higher INDIRECT X mentions scores than their DIRECT X mentions counterparts (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The Spearman\u0026rsquo;s rank correlation coefficient between DIRECT and INDIRECT X mentions was 0.53, suggesting a moderate positive correlation but no strict proportional relationship between the two metrics.\u003c/p\u003e \u003cp\u003eWhen comparing the combined score (DIRECT\u0026thinsp;+\u0026thinsp;INDIRECT X mentions) with the DIRECT X mentions, the minimum values remained almost unchanged \u0026mdash; increasing only slightly from 252 to 254 \u0026mdash; whereas the maximum value rose dramatically from 7,308 to 15,404, indicating a substantial expansion in the upper range (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the top 50 papers from the re-ranked list of all 100 papers based on their total combined scores. Among these 50 papers, 23 showed rank fluctuations of more than ten positions upward or downward. These substantial ranking shifts demonstrate that incorporating INDIRECT X mentions not only reveals the existence of second-order citations but fundamentally alters the evaluation of research impact, identifying influential papers that would otherwise be undervalued by conventional metrics relying solely on first-order citations.\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\u003eDescriptive statistics of 100 papers\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" 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=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eTotal Score\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003ein 1.5 years\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003eMin\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003e25th\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003ePercentile\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003e50th\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003ePercentile\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003e75th\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003ePercentile\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cem\u003eMax\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cem\u003eLarger than\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003ethe other\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eDIRECT X mentions\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e97,626\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e252\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e318.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e499\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e933.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e7,308\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e81\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eINDIRECT X mentions\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e70,567\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e629.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e8,625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTotal Score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e168,193\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e254\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e379\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e726\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e1,856\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e15,404\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e maps the 100 papers using INDIRECT X mentions on the x-axis and DIRECT X mentions on the y-axis, classifying them into four quadrants based on their mean values. Even though all target papers were among the top 100 in DIRECT X mentions, they were unevenly distributed across the four quadrants, with a majority (64 papers) concentrated in Q3, indicating that both DIRECT and INDIRECT X mentions exhibited a highly skewed distribution. This finding aligns with previous studies that have consistently reported highly skewed distributions in altmetric data (Banshal et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Our results extend these findings by demonstrating that not only first-order citations (DIRECT X mentions) but also second-order citations (INDIRECT X mentions) exhibit similarly skewed characteristics.\u003c/p\u003e \u003cp\u003eThe remaining papers were relatively well distributed across the other quadrants. Papers in Q1 generally showed high values in both metrics, representing research with strong dissemination both within and beyond academic communities, whereas Q2 papers exhibited high DIRECT but low INDIRECT X mentions, reflecting primarily academic attention. Among these, 12 papers in Q4 were newly identified as influential works not sufficiently captured by the conventional metric based on DIRECT X mentions. The increases in their scores were particularly striking \u0026mdash; as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, ten of these twelve papers exhibited substantial upward rank shifts, rising by approximately 25 to 37 positions, with two papers newly entering the top ten.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the paper titles classified into quadrants Q1, Q2, and Q4. While topics do not clearly separate across quadrants, certain thematic tendencies can be observed. Q1 papers frequently featured keywords related to practical applications in high-stakes domains (e.g., \"USMLE,\" \"Medical Education,\" \"Clinical Knowledge,\" \"Finance\"), evaluation of AI capabilities (e.g., \"Performance,\" \"Evaluating,\" \"theory of mind,\" \"Emergent Abilities\"), and societal implications (e.g., \"Labor Market Impact\"). Q2 papers tended to emphasize technical methodologies (e.g., \"deep reinforcement learning,\" \"Multimodal Deep Learning\"), domain-specific applications (e.g., \"protein sequences,\" \"Japanese Medical,\" \"soccer skills\"). Q4 papers exhibited more diverse characteristics, including technical documentation (e.g., \"GPT-4 Technical Report\"), workplace implications (e.g., \"Generative AI at Work\"), security and ethical concerns (e.g., \"Indirect Prompt Injection,\" \"Watermark,\" \"Moral Self-Correction\"), AI output evaluation (e.g., \"abstracts generated by ChatGPT,\" \"plagiarism detector\"), and interdisciplinary applications (e.g., \"ChemCrow,\" \"technosignatures,\" \"beat synchronization in rats\"). While Q2 predominantly featured technical and domain-specific topics, the tendency for Q4 to cluster papers related to real-world applications of AI suggests that INDIRECT X mentions capture discussions centered not within academic communication but within broader public and media discourse. This pattern indicates that second-order citations reflect reactions from non-academic audiences and highlight the societal dimensions of research influence. Particularly noteworthy is the placement of the \u003cem\u003e\u0026ldquo;GPT-4 Technical Report\u0026rdquo;\u003c/em\u003e in Q4. Despite being an OpenAI technical report outlining the foundational architecture of ChatGPT and having an exceptionally high citation count\u0026mdash;2,043 citations as of November 2025\u0026mdash;it was classified into Q4. While the papers with the highest citation counts, such as those published in \u003cem\u003ePLOS Digital Health\u003c/em\u003e and \u003cem\u003eNature\u003c/em\u003e, appear in Q1, the fact that the \u003cem\u003e\u0026ldquo;GPT-4 Technical Report\u0026rdquo;\u003c/em\u003e\u0026mdash;a globally influential and highly cited work\u0026mdash;fell into Q4 strongly underscores the necessity of measuring social dissemination through second-order citations. This finding serves as compelling evidence of the value and effectiveness of INDIRECT X mentions as an indicator for capturing societal-level engagement.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Time Series Analysis\u003c/h2\u003e \u003cp\u003eWhile the previous section examined the overall distribution and relative magnitude of DIRECT and INDIRECT X mentions, it remains unclear how these two indicators differ in their temporal diffusion patterns. Previous research has shown that Twitter mentions of scientific publications exhibit distinct temporal patterns (Priem \u0026amp; Costello, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Fang et al., 2020). However, these findings have been based exclusively on first-order citations (DIRECT X mentions in our terminology). The temporal dynamics of second-order citations (INDIRECT X mentions) remain unexplored. Understanding whether second-order citations emerge and evolve differently from first-order citations is essential for capturing the process of knowledge diffusion both within and beyond academia, as well as for accurately measuring the dynamics of research impact. Therefore, this study conducted a time-series analysis of \u003cem\u003eDIRECT X mentions\u003c/em\u003e and \u003cem\u003eINDIRECT X mentions\u003c/em\u003e to elucidate their respective characteristics in terms of immediacy and long-term influence.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eList of paper titles classified into quadrants Q1, Q2, and Q4\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003ePaper Title (Journal / Collection)\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003eCitation\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ1\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e1\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e2\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e3\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e5\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e7\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e8\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e9\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e11\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e12\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e14\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e16\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026ldquo;Sparks of Artificial General Intelligence: Early experiments with GPT-4\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Performance of ChatGPT on USMLE: Potential for AI-Assisted Medical Education Using Large Language Models\u0026rdquo; ( PLOS Digital Health )\u003c/p\u003e \u003cp\u003e\u0026ldquo;BloombergGPT: A Large Language Model for Finance\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Large Language Models Encode Clinical Knowledge\u0026rdquo; ( Nature )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Using deep learning to predict ideology from facial photographs: expressions, beauty, and extra-facial information\u0026rdquo; (Scientific Reports )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Human-level play in the game of Diplomacy by combining language models with strategic reasoning\u0026rdquo; ( Science )\u003c/p\u003e \u003cp\u003e\u0026ldquo;The connectome of an insect brain\u0026rdquo; ( Science (New York, N.Y.) )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Evaluating large language models in theory of mind tasks\u0026rdquo; ( Proceedings of the National Academy of Sciences of the United States of America )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Are Emergent Abilities of Large Language Models a Mirage?\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Dissociating language and thought in large language models\u0026rdquo; ( Trends in Cognitive Sciences )\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1,478\u003c/p\u003e \u003cp\u003e518\u003c/p\u003e \u003cp\u003e2,973\u003c/p\u003e \u003cp\u003e294\u003c/p\u003e \u003cp\u003e2,159\u003c/p\u003e \u003cp\u003e17\u003c/p\u003e \u003cp\u003e36\u003c/p\u003e \u003cp\u003e373\u003c/p\u003e \u003cp\u003e129\u003c/p\u003e \u003cp\u003e131\u003c/p\u003e \u003cp\u003e209\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eQ2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e6\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e15\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e17\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e18\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e19\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e22\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e24\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e26\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e28\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e33\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e35\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e37\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e38\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026ldquo;Learning agile soccer skills for a bipedal robot with deep reinforcement learning\u0026rdquo; ( Science Robotics )\u003c/p\u003e \u003cp\u003e\u0026ldquo;A Survey of Large Language Models\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Multimodal Deep Learning\u0026rdquo; ( International Conference on Machine Learning )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Emergent autonomous scientific research capabilities of large language models\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Superhuman artificial intelligence can improve human decision-making by increasing novelty\u0026rdquo; (Proceedings of the National Academy of Sciences of the United States of America )\u003c/p\u003e \u003cp\u003e\u0026ldquo;GPT detectors are biased against non-native English writers\u0026rdquo; (Patterns )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Talking About Large Language Models\u0026rdquo; ( Communications of the ACM )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Using AI to Implement Effective Teaching Strategies in Classrooms: Five Strategies, Including Prompts\u0026rdquo; ( SSRN Electronic Journal )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Large language models generate functional protein sequences across diverse families\u0026rdquo; (Nature Biotechnology )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Occupational Heterogeneity in Exposure to Generative AI\u0026rdquo; (SSRN Electronic Journal )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Evaluating GPT-4 and ChatGPT on Japanese Medical Licensing Examinations\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Assigning AI: Seven Approaches for Students, with Prompts\u0026rdquo; (SSRN Electronic Journal )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Foundation models for generalist medical artificial intelligence\u0026rdquo; ( Nature )\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e106\u003c/p\u003e \u003cp\u003e1,330\u003c/p\u003e \u003cp\u003e189\u003c/p\u003e \u003cp\u003e73\u003c/p\u003e \u003cp\u003e73\u003c/p\u003e \u003cp\u003e301\u003c/p\u003e \u003cp\u003e200\u003c/p\u003e \u003cp\u003e162\u003c/p\u003e \u003cp\u003e816\u003c/p\u003e \u003cp\u003e78\u003c/p\u003e \u003cp\u003e50\u003c/p\u003e \u003cp\u003e124\u003c/p\u003e \u003cp\u003e1,165\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eQ4\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e4\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e10\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e13\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e20\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e21\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e23\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e25\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e27\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e29\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e32\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e34\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e36\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026ldquo;GPT-4 Technical Report\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection\u0026rdquo; ( Proceedings of the 16th ACM Workshop on Artificial Intelligence and Security )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Generative AI at Work\u0026rdquo; ( \u003cem\u003eThe Quarterly Journal of Economics\u003c/em\u003e )\u003c/p\u003e \u003cp\u003e\u0026ldquo;A deep-learning search for technosignatures from 820 nearby stars\u0026rdquo; ( \u003cem\u003eNature Astronomy\u003c/em\u003e )\u003c/p\u003e \u003cp\u003e\u0026ldquo;ChemCrow: Augmenting large-language models with chemistry tools\u0026rdquo; (Nature Machine Intelligence )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Comparing scientific abstracts generated by ChatGPT to original abstracts using an artificial intelligence output detector, plagiarism detector, and blinded human reviewers\u0026rdquo; ( BioRxiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;A Watermark for Large Language Models\u0026rdquo; ( \u003cem\u003eInternational Conference on Machine Learning\u003c/em\u003e )\u003c/p\u003e \u003cp\u003e\u0026ldquo;The Capacity for Moral Self-Correction in Large Language Models\u0026rdquo; ( ArXiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;AI model GPT-3 (dis)informs us better than humans\u0026rdquo; ( \u003cem\u003eScience Advances\u003c/em\u003e )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Spontaneous beat synchronization in rats: Neural dynamics and motor entrainment\u0026rdquo; ( \u003cem\u003eScience Advances\u003c/em\u003e )\u003c/p\u003e \u003cp\u003e\u0026ldquo;Is GPT-3 all you need for low-data discovery in chemistry?\u0026rdquo; ( ChemRxiv )\u003c/p\u003e \u003cp\u003e\u0026ldquo;The growing influence of industry in AI research\u0026rdquo; ( \u003cem\u003eScience\u003c/em\u003e )\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2,043\u003c/p\u003e \u003cp\u003e190\u003c/p\u003e \u003cp\u003e673\u003c/p\u003e \u003cp\u003e34\u003c/p\u003e \u003cp\u003e336\u003c/p\u003e \u003cp\u003e387\u003c/p\u003e \u003cp\u003e110\u003c/p\u003e \u003cp\u003e44\u003c/p\u003e \u003cp\u003e170\u003c/p\u003e \u003cp\u003e28\u003c/p\u003e \u003cp\u003e26\u003c/p\u003e \u003cp\u003e131\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eCitation counts obtained via OpenAlex (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://openalex.org/\u003c/span\u003e\u003cspan address=\"https://openalex.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) as of November 13, 2025.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Verification of immediacy\u003c/h2\u003e \u003cp\u003eAccording to Priem and Costello (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) 15% of first-order citations on X occur on the day of publication, and 40% within the first week. In this study, the reaction appeared even more rapidly: as shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, approximately 30% of DIRECT X mentions occurred on the publication day, and about 90% within the first week. In contrast, INDIRECT X mentions showed a slower response, with 15.8% appearing on the publication day and 39.5% within the first week. Consequently, the overall proportions of INDIRECT X mentions were close to the benchmark values reported by Priem and Costello (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). Considering that traditional citation metrics typically require about a year to accumulate measurable counts, these results indicate that INDIRECT X mentions still exhibit considerable immediacy.\u003c/p\u003e \u003cp\u003eFurther analysis of the response timing for each paper revealed that in 77% of the cases, DIRECT X mentions appeared earlier than INDIRECT X mentions, while no papers exhibited earlier INDIRECT X mentions. The average difference in response time between the two indicators was 64.0 days (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eElapsed days from publication to first post\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eDIRECT X mentions\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003eINDIRECT X mentions *exclude 0\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eProportion on the day of publication\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15.8%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eProportion within one week after publication\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e90.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e39.5%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eDIRECT X mentions is faster than INDIRECT\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e57 papers\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eDIRECT X mentions is as same as INDIRECT\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e19 papers\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eDIRECT X mentions is later than INDIRECT\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0 papers\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eMean difference in elapsed days (INDIRECT - DIRECT)\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e64.0 days\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAll items except \u0026ldquo;Proportion on the day of publication\u0026rdquo; and \u0026ldquo;Proportion within one week after publication\u0026rdquo; for DIRECT X mentions exclude papers with an INDIRECT X mentions value of zero, which applies to 24 out of the 100 papers.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Verification of long-term influence\u003c/h2\u003e \u003cp\u003eTo determine whether first- and second-order citations differ in the persistence of their influence, we performed a longitudinal analysis of each paper. We calculated the number of days required for each paper to reach between 50% and 100% of its final score. Previous research by Fang et al. (2020) found that Twitter data accumulate very rapidly, with a half-life of only 13 days across multiple scientific fields, indicating that Twitter is among the fastest altmetric data sources in accumulating mentions shortly after publication. However, this finding was based on only first-order citations (DIRECT X mentions)\u003c/p\u003e \u003cp\u003eOur results reveal a consistent pattern of delayed dissemination for INDIRECT X mentions across all thresholds (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). At the 50% threshold, DIRECT X mentions required only 7.0 days (median) or 42.5 days (mean), whereas INDIRECT X mentions required substantially longer periods\u0026mdash;41.5 days (median) and 103.9 days (mean). This gap widened progressively through the 60%, 70%, and 80% thresholds, demonstrating that second-order citations accumulated more slowly at each stage.\u003c/p\u003e \u003cp\u003eBy the results show that \u003cem\u003eDIRECT X mentions\u003c/em\u003e reached 90% of their cumulative score after an average of 125.4 days (median\u0026thinsp;=\u0026thinsp;66.5 days), whereas \u003cem\u003eINDIRECT X mentions\u003c/em\u003e required a longer period, with an average of 188.4 days (median\u0026thinsp;=\u0026thinsp;127 days) (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Across the 50\u0026ndash;90% thresholds, INDIRECT X mentions consistently appeared later than DIRECT X mentions, reflecting the delayed and secondary nature of public dissemination through intermediary media. However, at the 100% threshold, DIRECT X mentions exhibited a longer tail, suggesting that academic users occasionally revisited or re-shared the papers long after their initial release.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe number of days required for each paper to reach between 50% and 100% of its final score\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\" morerows=\"1\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e\u003cem\u003eDIRECT X mentions (days)\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c6\" namest=\"c4\"\u003e \u003cp\u003e\u003cem\u003eINDIRECT X mentions (days)\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eMedian\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e\u003cem\u003eMean\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eMedian\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003eMean\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e50%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e42.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e41.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e103.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e60%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e46.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e52.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e113.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e70%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e12.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e59.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e76.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e127.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e80%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e31.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e78.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e89.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e152.3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e90%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e66.5\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e\u003cb\u003e125.4\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e127.0\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e188.4\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e100%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e502.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e460.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e378.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e297.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eFigure\u0026nbsp;3\u003c/b\u003e The distribution of papers based on the difference in days to the 90% attainment (x-axis) and the difference in cumulative score between \u003cem\u003eDIRECT\u003c/em\u003e and \u003cem\u003eINDIRECT X mentions\u003c/em\u003e (y-axis). The numbers in each of the four quadrants, labeled A, B, C, and D, represent the number of papers distributed within each area, based on the axes where x\u0026thinsp;=\u0026thinsp;0 and y\u0026thinsp;=\u0026thinsp;0. Since the 24 papers with an INDIRECT X mentions value of zero are always classified into quadrant C, they are highlighted in orange and counted separately.\u003c/p\u003e \u003cp\u003eFigure 3 illustrates the distribution of papers based on the difference in elapsed days (x-axis) and the difference in cumulative values (y-axis) between \u003cem\u003eDIRECT\u003c/em\u003e and \u003cem\u003eINDIRECT X mentions\u003c/em\u003e at the 90% threshold. The results reveal that in more than 85% of cases, \u003cem\u003eINDIRECT X mentions\u003c/em\u003e required a longer time to converge, regardless of their absolute magnitude relative to \u003cem\u003eDIRECT X mentions\u003c/em\u003e.\u003c/p\u003e \u003cp\u003eThese findings suggest that even after first-order citations plateau, second-order citations continue to contribute to the sustained social visibility of a paper, highlighting the crucial role of indirect information diffusion in \u003cem\u003ereactivating\u003c/em\u003e or \u003cem\u003esustaining\u003c/em\u003e the public impact of research over time.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Network Analysis\u003c/h2\u003e \u003cp\u003eWhile the previous section revealed differences in the speed and persistence of dissemination between first- and second-order citations, it remains unclear who actually participates in these dissemination processes.\u003c/p\u003e \u003cp\u003eUnderstanding whether the posts are made primarily by researchers or by broader non-academic audiences is crucial for evaluating research impact. In this chapter, we examine the possibility that second-order citations exert social influence within different audience spheres compared to first-order citations.\u003c/p\u003e \u003cp\u003eSpecifically, based on the actual post data of DIRECT X mentions and INDIRECT X mentions, we analyze user overlap and the proportion of researchers by constructing user networks consisting of posting users and mentioned users (i.e., users referenced through mentions, replies, or quotes).\u003c/p\u003e \u003cp\u003eFirst, to assess the degree of network overlap between the two groups, the mean Jaccard index for each paper was calculated. The results revealed a remarkably low level of user overlap, with the average Jaccard index for posting users was 0.012, for mentioned users was 0.034, and for all users combined was 0.017. This indicates that the users who directly posted links to papers (first-order citations) and those who shared intermediary webpages (second-order citations) were almost entirely distinct. This tendency remained consistent across all four quadrants (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eNext, we calculated the proportion of researcher users in both the first-order and second-order citation groups. Among posting users, Group 1 (DIRECT X mentions) comprised 9.9% researchers, whereas Group 2 (INDIRECT X mentions) comprised only 4.1%, approximately half the proportion of Group 1. This result indicates that first-order citations are primarily driven by researchers, while second-order citations are more strongly associated with non-academic communicators. Differences were also observed across quadrants. Quadrant 2, characterized by high DIRECT X mentions and low INDIRECT X mentions, exhibited the highest share of researchers at 10.8%, approximately twice that of Quadrant 4 (5.4%), which displayed the opposite pattern (low DIRECT and high INDIRECT X mentions). In Quadrant 4, however, the proportion of researchers within both Groups 1 and 2 reached the highest values\u0026mdash;16.2% and 7.1%, respectively. Regarding mentioned users, although Quadrant 2 showed particularly high values, the overall proportions of researchers across both groups were generally comparable. Compared with posting users, Group 2 contained a relatively higher proportion of researcher accounts, yet there was almost no overlap between researcher users in Group 1 and Group 2 (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e presents the top 20 nodes with the highest frequencies of occurrence among posting and mentioned users, summarizing the representative account classifications for both Group 1 and Group 2. The results indicate that the posting activity in Group 1 was influenced by the bot account \u0026ldquo;arxivabs\u0026rdquo;, which automatically replies with corresponding paper links. In addition, many of the top posting users in Group 1 were identified as researchers, most of whom were the authors of the referenced papers. In contrast, Group 2 included three researcher accounts; however, the majority consisted of general users and multiple news media sites, none of which were paper authors or media authors. Regarding the mentioned users, Group 1 was predominantly composed of researcher accounts, most of whom were again paper authors. The remaining top-mentioned users primarily consisted of research-related accounts, including research and development organizations (e.g., \u003cem\u003e@openai, @penn\u003c/em\u003e) and academic publication sites (e.g., \u003cem\u003e@nature, @arxiv\u003c/em\u003e). In contrast, the top-mentioned users in Group 2 were largely news outlets, with only one researcher account identified \u0026mdash; a media article author. Several general users were also found to be authors of media articles, further illustrating the distinctive composition of Group 2\u0026rsquo;s dissemination network.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eProportion of researcher among all users\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eTotal Users\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003eResearcher\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003e(overlapping researcher)\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003ePercentage of researchers\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003e1 \u003cem\u003ePositing Users\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eGroup1\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2,512\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e249\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9.9%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eGroup2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9,349\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e386\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ1\u003c/em\u003e\u003c/p\u003e \u003cp\u003eGroup1\u003c/p\u003e \u003cp\u003eGroup2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5,055\u003c/p\u003e \u003cp\u003e1,336\u003c/p\u003e \u003cp\u003e3,791\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e203 (5)\u003c/p\u003e \u003cp\u003e101\u003c/p\u003e \u003cp\u003e107\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.0%\u003c/p\u003e \u003cp\u003e7.6%\u003c/p\u003e \u003cp\u003e2.8%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ2\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup1\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e713\u003c/p\u003e \u003cp\u003e333\u003c/p\u003e \u003cp\u003e395\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e77 (5)\u003c/p\u003e \u003cp\u003e54\u003c/p\u003e \u003cp\u003e28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10.8%\u003c/p\u003e \u003cp\u003e16.2%\u003c/p\u003e \u003cp\u003e7.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ3\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup1\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e845\u003c/p\u003e \u003cp\u003e265\u003c/p\u003e \u003cp\u003e593\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e48 (3)\u003c/p\u003e \u003cp\u003e34\u003c/p\u003e \u003cp\u003e17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5.7%\u003c/p\u003e \u003cp\u003e12.8%\u003c/p\u003e \u003cp\u003e2.9%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ4\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup1\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6,470\u003c/p\u003e \u003cp\u003e793\u003c/p\u003e \u003cp\u003e5,800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e348 (13)\u003c/p\u003e \u003cp\u003e83\u003c/p\u003e \u003cp\u003e278\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5.4%\u003c/p\u003e \u003cp\u003e10.5%\u003c/p\u003e \u003cp\u003e4.8%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003e2 \u003cem\u003eReferenced Users\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eGroup1\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,855\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e176\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9.5%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eGroup2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,694\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e141\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8.3%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ1\u003c/em\u003e\u003c/p\u003e \u003cp\u003eGroup1\u003c/p\u003e \u003cp\u003eGroup2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,605\u003c/p\u003e \u003cp\u003e1,051\u003c/p\u003e \u003cp\u003e600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e113 (13)\u003c/p\u003e \u003cp\u003e80\u003c/p\u003e \u003cp\u003e46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7.0%\u003c/p\u003e \u003cp\u003e7.6%\u003c/p\u003e \u003cp\u003e7.7%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ2\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup1\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e262\u003c/p\u003e \u003cp\u003e168\u003c/p\u003e \u003cp\u003e101\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47 (2)\u003c/p\u003e \u003cp\u003e42\u003c/p\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17.9%\u003c/p\u003e \u003cp\u003e25.0%\u003c/p\u003e \u003cp\u003e6.9%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ3\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup1\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e250\u003c/p\u003e \u003cp\u003e119\u003c/p\u003e \u003cp\u003e141\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e25 (3)\u003c/p\u003e \u003cp\u003e17\u003c/p\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10.0%\u003c/p\u003e \u003cp\u003e14.3%\u003c/p\u003e \u003cp\u003e7.8%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQ4\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup1\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eGroup2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,611\u003c/p\u003e \u003cp\u003e572\u003c/p\u003e \u003cp\u003e1,094\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e146 (10)\u003c/p\u003e \u003cp\u003e55\u003c/p\u003e \u003cp\u003e101\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9.1%\u003c/p\u003e \u003cp\u003e9.6%\u003c/p\u003e \u003cp\u003e9.2%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe numbers in parentheses for Q1\u0026ndash;Q4 indicate the counts of overlapping researcher users between Group1 and Group2.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis study proposed INDIRECT X mentions, a novel indicator based on second-order citations, to complement traditional DIRECT X mentions, which capture only first-order dissemination of academic papers. The analysis demonstrated that, consistent with previous research, second-order citations are also widespread in the AI domain. Despite selecting the top 100 papers based on DIRECT X mentions, the volume of INDIRECT X mentions was found to be of comparable magnitude. When combining both indicators into a total score, the overall distribution expanded significantly, allowing papers that achieved high levels of societal attention to be more clearly distinguished.\u003c/p\u003e \u003cp\u003eThe time-series analysis revealed that the two indicators exhibit distinct diffusion characteristics. Specifically, INDIRECT X mentions showed slower initial reactions but a greater tendency for long-term dissemination over time. These differences likely stem from the structural mechanism of data aggregation. Since INDIRECT X mentions can only be counted once intermediary webpages\u0026mdash;such as news articles or blogs\u0026mdash;have been created, they tend to appear later but may continue to spread as long as new intermediary webpages emerge. In contrast, DIRECT X mentions reflected an immediate surge in attention, particularly at key events such as the release of a paper\u0026rsquo;s second edition, where a \u0026ldquo;reactivation\u0026rdquo; phenomenon was observed (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003e). These findings suggest that indirect dissemination via intermediary webpages functions as a mechanism for sustaining public engagement with research beyond the initial citation period. Consequently, INDIRECT X mentions can be interpreted as an indicator that captures sustained societal engagement, a dimension previously overlooked by conventional altmetric measures. The network analysis further revealed that first- and second-order citations are generated by largely distinct groups of users. Compared to first-order citations, second-order citations involved substantially fewer researchers. This pattern mirrors the findings of Alperin et al. (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) in their COVID-19 case study, confirming that such a distinction is a consistent feature of second-order dissemination. Examination of the top posting and mentioned users indicated that DIRECT X mentions were primarily generated by bots (e.g., \u003cem\u003earxivabs\u003c/em\u003e) and paper authors, whereas INDIRECT X mentions were dominated by general users, journalists, and news media accounts with limited direct involvement in research. The difference in mentioned-user composition likely reflects posting conventions on X, where users often credit the original source\u0026mdash;either a paper or a media article\u0026mdash;by tagging its hosting site or author account. Given that users on X form distinct follower networks, variations in posting-user composition lead to markedly different diffusion pathways. Thus, both DIRECT and INDIRECT X mentions, which incorporate repost and quote activity, represent audience-specific patterns of engagement with research outputs.\u003c/p\u003e \u003cp\u003eThis interpretation is further supported by the quadrant-based analysis. Papers in Quadrant 2 (Q2), characterized by high DIRECT and low INDIRECT X mentions, had the highest proportion of researchers (10.8%), whereas Quadrant 4 (Q4), showing the opposite pattern, exhibited roughly half that rate (5.4%). Q2 primarily comprised technical studies such as \u003cem\u003e\u0026ldquo;Learning Agile Soccer Skills for a Bipedal Robot with Deep Reinforcement Learning\u0026rdquo;\u003c/em\u003e and \u003cem\u003e\u0026ldquo;Multimodal Deep Learning\u0026rdquo;\u003c/em\u003e, while Q4 included socially and ethically oriented papers such as \u003cem\u003e\u0026ldquo;GPT-4 Technical Report\u0026rdquo;\u003c/em\u003e, \u003cem\u003e\u0026ldquo;Generative AI at Work\u0026rdquo;\u003c/em\u003e, and \u003cem\u003e\u0026ldquo;The Capacity for Moral Self-Correction in Large Language Models.\u0026rdquo;\u003c/em\u003e These results suggest that INDIRECT X mentions more effectively capture the societal dimension of research impact that traditional altmetrics often overlook.\u003c/p\u003e \u003cp\u003eThe fact that the \u003cem\u003e\u0026ldquo;GPT-4 Technical Report,\u0026rdquo;\u003c/em\u003e which had been cited 2,043 times as of November 2025, was classified into Q4\u0026mdash;despite its exceptionally high scholarly influence and relatively low DIRECT X mentions\u0026mdash;strongly reinforces the necessity of detecting the societal impact generated through second-order citations. This finding serves as a clear demonstration of the effectiveness of INDIRECT X mentions. Moreover, it suggests that research fields attracting broad public attention may exhibit distinctive dissemination patterns that differ from those observed in more traditional academic domains. Papers highly ranked on both indicators \u0026mdash;such as \u003cem\u003e\u0026ldquo;Sparks of Artificial General Intelligence\u0026rdquo;\u003c/em\u003e, \u003cem\u003e\u0026ldquo;GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models\u0026rdquo;\u003c/em\u003e, and \u003cem\u003e\u0026ldquo;Performance of ChatGPT on USMLE: Potential for AI-Assisted Medical Education Using Large Language Models\u0026rdquo;\u003c/em\u003e\u0026mdash;represent research that successfully engaged both academic and public audiences.\u003c/p\u003e \u003cp\u003eNevertheless, this study is limited by its focus on the top 100 papers in the AI domain. Further work is required to assess cross-disciplinary reproducibility and to establish a generalized classification framework that integrates both indicators. Future research should also analyze the content of intermediary webpages to better understand how academic findings are reframed and disseminated in social and media contexts, thereby enabling a more comprehensive evaluation of the societal influence of research.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study conceptualized and operationalized \u003cem\u003esecond-order citations\u003c/em\u003e as INDIRECT X mentions, and conducted a comparative validation against the existing first-order citation\u0026ndash;based indicator, DIRECT X mentions, which aggregates the number and dissemination counts of direct posts linking to academic papers. Our findings reveal that in the AI domain, a substantial amount of indirect dissemination occurs through news articles and blogs, involving user groups that more closely represent general audiences. Furthermore, second-order citations exhibited a tendency to sustain engagement on X over a longer period, thereby providing clearer evidence of their importance for accurately capturing the dynamics of research dissemination on social media. While this study is limited to a case analysis of 100 AI-related papers, our initial hypothesis\u0026mdash;that understanding the true dissemination of academic outputs requires accounting not only for direct but also indirect citations\u0026mdash;was supported within this scope. In rapidly evolving and socially visible research fields such as AI, where both academic and non-academic interest is particularly high, the altmetrics initiative of \u0026ldquo;visualizing the real-time societal impact of research\u0026rdquo; holds significant value. In this respect, this study provides meaningful insights by validating the visualization and applicability of both indicators in the AI field. Although further empirical validation is needed for the proposed INDIRECT X mentions, this study contributes by establishing its definition as a measurable indicator and by presenting a reproducible and transparent aggregation framework based on the existing Altmetrics\u0026rsquo;s \u003cem\u003eMentions\u003c/em\u003e data. The proposed procedure can be replicated using publicly accessible data categories (News and Blogs) and standard APIs, ensuring methodological consistency and facilitating future comparative studies.\u003c/p\u003e \u003cp\u003ePrevious research has shown that intermediary webpages provided by data suppliers exhibit language-based bias (Ortega, 2019b). A similar pattern was observed in the present dataset: both indicators were strongly skewed toward English, and particularly for Japanese, a substantial discrepancy was evident between the proportions of DIRECT X mentions and INDIRECT X mentions. Furthermore, 18 out of the 100 papers had zero intermediary webpages, suggesting the need for careful validation of the dataset\u0026rsquo;s reliability. Given that the overlap of intermediary webpage coverage across providers is known to be low, the use of multiple data providers is recommended (Haustein et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2013\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe two indicators proposed in this study\u0026mdash;DIRECT X mentions and INDIRECT X mentions\u0026mdash;hold considerable potential for future applications. Although further investigation is required to establish standardized methods for paper classification, the differences in the intrinsic characteristics of these indicators suggest promising opportunities for developing classification frameworks that leverage both metrics. Existing research has extensively examined the relationship between DIRECT X mentions and conventional indicators of academic success, such as citation counts, yet no clear or consistent correlation has been confirmed (Haustein et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Bornmann, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Mazurek et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Howoldt et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In contrast, INDIRECT X mentions may represent a metric that captures a distinct dimension of societal influence not reflected in traditional indicators, thereby offering substantial potential as a predictive measure. Both indicators demonstrated convergence earlier than 1.5 years after publication\u0026mdash;a notably shorter timescale than that of conventional citations, which typically begin accumulating after one year or more. Furthermore, the case of the \u0026ldquo;GPT-4 Technical Report\u0026rdquo; in the Q4 classification revealed the presence of large-scale public discourse occurring on X that was not captured by first-order\u0026ndash;based indicators. This finding underscores the significant role of INDIRECT X mentions in reflecting public engagement within the broader context of citizen participation in science and highlights important directions for future research.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAi Kishimoto wrote the main manuscript text and prepared all Figures and Tables. All authors reviewed the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll data supporting the findings of this study are available from National Graduate Institute for Policy Studies.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAlhoori, H., \u0026amp; Furuta, R. (2014). Do Altmetrics Follow the Crowd or Does the Crowd Follow Altmetrics? \u003cem\u003eIEEE Press\u003c/em\u003e, 375\u0026ndash;378.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlperin, J. P., Fleerackers, A., Riedlinger, M., \u0026amp; Haustein, S. (2024). Second-order citations in altmetrics: A case study analyzing the audiences of COVID-19 research in the news and on social media. \u003cem\u003eQuantitative Science Studies\u003c/em\u003e, 5(2), 366\u0026ndash;382. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1162/qss_a_00298\u003c/span\u003e\u003cspan address=\"10.1162/qss_a_00298\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBanshal, S. K., Gupta, S., Lathabai, H. H., \u0026amp; Singh, V. K. (2022). Power laws in altmetrics: An empirical analysis. \u003cem\u003eJournal of Informetrics\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e, 101309. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.joi.2022.101309\u003c/span\u003e\u003cspan address=\"10.1016/j.joi.2022.101309\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBornmann, L. (2015). Alternative metrics in scientometrics: A meta-analysis of research into three altmetrics. \u003cem\u003eScientometrics\u003c/em\u003e, \u003cem\u003e103\u003c/em\u003e(3), 1123\u0026ndash;1144.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBornmann, L., \u0026amp; Haunschild, R. (2019). Societal impact measurement of research papers. In W. Gl\u0026auml;nzel, H. F. Moed, U. Schmoch, \u0026amp; M. Thelwall (Eds.), S\u003cem\u003epringer handbook of science and technology indicators\u003c/em\u003e (pp. 609\u0026ndash;632). Springer Nature.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrody, T., Harnad, S., \u0026amp; Carr, L. (2006). Earlier Web usage statistics as predictors of later citation impact. \u003cem\u003eJournal of the American Society for Information Science and Technology\u003c/em\u003e, 57(8), 1060\u0026ndash;1072. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/asi.20373\u003c/span\u003e\u003cspan address=\"10.1002/asi.20373\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCostas, R., Mongeon, P., Ferreira, M. R., van Honk, J., \u0026amp; Franssen, T. (2020). Large-scale identification and characterization of scholars on Twitter. \u003cem\u003eQuantitative Science Studies\u003c/em\u003e, 1(2), 771\u0026ndash;791. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1162/qss_a_00047\u003c/span\u003e\u003cspan address=\"10.1162/qss_a_00047\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFang, Z., \u0026amp; Costas, R. (2020). Studying the accumulation velocity of altmetric data tracked by Altmetric.com. \u003cem\u003eScientometrics\u003c/em\u003e, \u003cem\u003e123\u003c/em\u003e, 1077\u0026ndash;1101. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11192-020-03405-9\u003c/span\u003e\u003cspan address=\"10.1007/s11192-020-03405-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGingras, Y. (2014). Criteria for evaluating indicators. In B. Cronin \u0026amp; C. R. Sugimoto (Eds.), \u003cem\u003eBeyond bibliometrics: Harnessing multidimensional indicators of scholarly impact\u003c/em\u003e (pp. 109\u0026ndash;125). MIT Press.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaustein, S. (2019). Scholarly twitter metrics. In \u003cem\u003eSpringer handbook of science and technology indicators\u003c/em\u003e (pp. 729\u0026ndash;760). Cham: Springer International Publishing.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaustein, S., Peters, I., Sugimoto, C. R., Thelwall, M., \u0026amp; Larivi\u0026egrave;re, V. (2013) Tweeting biomedicine: an analysis of tweets and citations in the biomedical literature. \u003cem\u003eJournal of the Association for Information Science and Technology\u003c/em\u003e, \u003cem\u003e65\u003c/em\u003e(4), 656\u0026ndash;669. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/asi.23101\u003c/span\u003e\u003cspan address=\"10.1002/asi.23101\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHowoldt, D., Kroll, H., Neuh\u0026auml;usler, P., \u0026amp; Feidenheimer, A. (2023). Understanding researchers\u0026rsquo; Twitter uptake, activity and popularity\u0026mdash;an analysis of applied research in Germany. \u003cem\u003eScientometrics\u003c/em\u003e, \u003cem\u003e128\u003c/em\u003e(1), 325\u0026ndash;344.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKe, Q., Ahn, Y., \u0026amp; Sugimoto, C. R.(2017). A systematic identification and analysis of scientists on Twitter. \u003cem\u003ePLoS ONE\u003c/em\u003e 12(4): e0175368. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0175368\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0175368\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMazurek, G., Gorska, A., Korzynski, P., \u0026amp; Silva, S. (2022). Social networking sites and researcher\u0026rsquo;s success. \u003cem\u003eJournal of Computer Information Systems\u003c/em\u003e, \u003cem\u003e62\u003c/em\u003e(2), 259\u0026ndash;266.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMongeon, P., Bowman, T., \u0026amp; Costas, R. (2022). Open dataset of scholars on Twitter. \u003cem\u003eZenodo\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5281/zenodo.7013518\u003c/span\u003e\u003cspan address=\"10.5281/zenodo.7013518\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMongeon, P., Bowman, T. D., \u0026amp; Costas, R. (2023). An open data set of scholars on Twitter. \u003cem\u003eQuantitative Science Studies\u003c/em\u003e, 4(2), 314\u0026ndash;324. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1162/qss_a_00250\u003c/span\u003e\u003cspan address=\"10.1162/qss_a_00250\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOrtega,J. L. (2018). Reliability and accuracy of altmetric providers: a comparison among Altmetric.com, PlumX and Crossref Event Data.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOrtega, J. L. (2020a). Altmetrics data providers: A meta-analysis review of the coverage of metrics and publication. \u003cem\u003eEl Profesional de La Informaci\u0026oacute;n\u003c/em\u003e, 29(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3145/epi.2020.ene.07\u003c/span\u003e\u003cspan address=\"10.3145/epi.2020.ene.07\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOrtega,J. L. (2020b). Blogs and news sources coverage in altmetrics data providers: a comparative analysis by country, language, and subject. \u003cem\u003eScientometrics\u003c/em\u003e, 122, 555\u0026ndash;572. Springer Nature.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePriem, J., \u0026amp; Costello, K. L. (2010). How and why scholars cite on Twitter. \u003cem\u003eProceedings of the American Society for Information Science and Technology, 47\u003c/em\u003e(1), 1\u0026ndash;4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/meet.14504701201\u003c/span\u003e\u003cspan address=\"10.1002/meet.14504701201\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePriem, J., Taraborelli,D., Groth, D., \u0026amp; Neylon,C.(2010). Altmetrics: A manifesto.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePriem, J. (2014). Altmetrics. In B. Cronin \u0026amp; C. R. Sugimoto (Eds.), \u003cem\u003eBeyond bibliometrics: Harnessing multidimensional indicators of scholarly impact\u003c/em\u003e (pp. 263\u0026ndash;288). MIT Press.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, H., Xiao, T., Xu, S., \u0026amp; Wang, Y. (2019). Who posts scientific tweets? An investigation into the productivity, locations, and identities of scientific tweeters. Journal of Informetrics, 13(3), 841\u0026ndash;855.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Raan, A. F. J. (2019). Measuring science: Basic principles and application of advanced bibliometrics. In W. Gl\u0026auml;nzel, H. F. Moed, U. Schmoch, \u0026amp; M. Thelwall (Eds.), Springer handbook of science and technology indicators (pp. 237\u0026ndash;280). Springer Nature.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWouters, P. (2014). The citation: From culture to infrastructure. In B. Cronin \u0026amp; C. R. Sugimoto (Eds.), \u003cem\u003eBeyond bibliometrics: Harnessing multidimensional indicators of scholarly impact\u003c/em\u003e (pp. 47\u0026ndash;66). MIT Press.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWouters, P., Zahedi, Z., \u0026amp; Costas, R. (2019). Social media metrics for new research evaluation.In W. Gl\u0026auml;nzel, H. F. Moed, U. Schmoch, \u0026amp; M. Thelwall (Eds.), Springer handbook of science and technology indicators (pp. 687\u0026ndash;713). Springer Nature.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZahedi, Z., Costas, R. (2018). General discussion of data quality challenges in social media metrics: Extensive comparison of four major altmetric data aggregators. \u003cem\u003ePloS one\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(5), e0197326. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0197326\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0197326\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, L., Gou, Z., Fang, Z., Sivertsen, G., \u0026amp; Huang, Y. (2023). Who tweets scientific publications? A large-scale study of tweeting audiences in all areas of research. Journal of the Association for Information Science and Technology, 74, 1485\u0026ndash;1497. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/asi.24830\u003c/span\u003e\u003cspan address=\"10.1002/asi.24830\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col start=3\u003e\u003cli\u003e\u003cspan\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.altmetric.com/\u003c/span\u003e\u003cspan address=\"https://www.altmetric.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.elsevier.com/insights/metrics/plumx\u003c/span\u003e\u003cspan address=\"https://www.elsevier.com/insights/metrics/plumx\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://profiles.impactstory.org/\u003c/span\u003e\u003cspan address=\"https://profiles.impactstory.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Altmetric.com, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.altmetric.com/blog/200-million-mentions-altmetric-has-now-officially-tracked-over-200-million-mentions-of-nearly-20-million-research-outputs/\u003c/span\u003e\u003cspan address=\"https://www.altmetric.com/blog/200-million-mentions-altmetric-has-now-officially-tracked-over-200-million-mentions-of-nearly-20-million-research-outputs/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Altmetric.com - X (formerly Twitter)\u003c/span\u003e\u003cdiv id=\"Par11\" class=\"Para\"\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://help.altmetric.com/support/solutions/articles/6000235926-x-formerly-twitter-\u003c/span\u003e\u003cspan address=\"https://help.altmetric.com/support/solutions/articles/6000235926-x-formerly-twitter-\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/div\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://chatgpt.com/\u003c/span\u003e\u003cspan address=\"https://chatgpt.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://nodexl.com/\u003c/span\u003e\u003cspan address=\"https://nodexl.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":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":"Altmetrics, INDIRECT X mensions, Twitter, social impact, second-order citations","lastPublishedDoi":"10.21203/rs.3.rs-8121384/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8121384/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study reexamines the potential of altmetrics as indicators for evaluating the societal impact of research, focusing specifically on X (formerly Twitter). Existing X-based metrics primarily measure \u003cem\u003efirst-order citations\u003c/em\u003e\u0026mdash;posts that directly link to academic papers. However, such metrics overlook \u003cem\u003esecond-order citations\u003c/em\u003e, which disseminate research indirectly through intermediary webpages such as news articles and blogs. To address this gap, we propose a new indicator, INDIRECT X mentions, designed to quantify the indirect dissemination of scholarly outputs. Using data from the artificial intelligence (AI) domain, this study compares the performance of DIRECT and INDIRECT X mentions over a 1.5-year period following publication. The results reveal that second-order citations occur at a comparable scale to first-order citations and exhibit distinctive diffusion characteristics: while INDIRECT X mentions demonstrate lower immediacy, they sustain engagement over a longer period. Furthermore, network analysis shows that the two metrics are generated by largely non-overlapping user groups, with INDIRECT X mentions associated more strongly with non-academic and media users. These findings suggest that INDIRECT X mentions provide a complementary perspective for measuring how academic knowledge circulates and resonates beyond scholarly communities, offering a more comprehensive framework for assessing the societal reach of research.\u003c/p\u003e","manuscriptTitle":"Proposal of INDIRECT X Mentions as an Altmetrics Indicator: Dissemination of Research Papers on X via Web News and Blogs","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-18 18:41:29","doi":"10.21203/rs.3.rs-8121384/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":"414dbb84-d969-4bef-8d24-77a864339ea7","owner":[],"postedDate":"December 18th, 2025","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-19T20:36:47+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-06T14:21:30+00:00","index":24,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-19T20:39:27+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-18 18:41:29","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8121384","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8121384","identity":"rs-8121384","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}