learning analytics

This study addresses the lack of a systematic approach for analyzing how tutor behaviors influence learning outcomes in one-on-one tutoring. It proposes the first structured classification framework that categorizes tutoring behaviors into four support types, innovatively introducing a “student engagement spectrum” to differentiate between guided reasoning and direct explanation. Grounded in cognitive and learning sciences, the framework was iteratively refined through a hybrid deductive–inductive methodology applied to transcribed real-world tutoring dialogues. Designed to support both expert coding and large-scale structured annotation, this framework provides a scalable foundation for AI-driven automated behavior recognition, computational modeling, and empirical investigations into the relationship between instructional behaviors and learning gains.

This study addresses the lack of systematic preprocessing standards, integrated analytical workflows, and cross-method consistency checks in current computer-based assessment process data. To bridge this gap, the authors propose an end-to-end analytical framework featuring a unified preprocessing pipeline and a dual-path analysis paradigm that synergistically combines feature engineering with model-based inference. The framework incorporates large language models (LLMs) to standardize action sequences and facilitate process-data-driven differential item functioning (DIF) detection. Technically, it integrates timestamp correction, action chunking, n-gram and TF-IDF feature extraction, multidimensional scaling, hidden Markov modeling, and subtask identification. Empirical results demonstrate that n-gram–based behavioral clustering offers diagnostic value for incorrect responders, multidimensional scaling effectively reconstructs behavioral constructs, and process data can identify and mitigate construct-irrelevant group differences.

This study investigates how K–12 mathematics teachers allocate limited real-time instructional support based on students’ prior help-seeking history and current engagement, and evaluates the cross-lesson learning effects of such support. Integrating teacher interviews with large-scale interaction data from the MATHia intelligent tutoring system, the research employs mixed-effects models, cross-lagged panel analysis, and additive factor models to reveal a “stickiness” in teacher attention: students who previously received help are more likely to be supported again. Although this targeted intervention significantly enhances immediate lesson performance, it shows no significant predictive effect on skill mastery in subsequent lessons. These findings offer empirical evidence for optimizing teacher attention allocation and advancing educational equity.

Intelligent tutoring systems (ITS) in curriculum-based online learning risk exacerbating academic achievement gaps among students. Method: This paper proposes CTGraph, the first self-supervised graph representation learning framework that explicitly incorporates curriculum structure priors to construct student behavior graphs—modeling multidimensional learning signals including learning pathways, content coverage, engagement intensity, and conceptual mastery. A graph neural network performs graph-level encoding to enable cross-cohort behavioral comparison and stage-wise difficulty localization. Contribution/Results: Experiments demonstrate that CTGraph accurately identifies at-risk students, pinpoints optimal intervention timing with fine-grained temporal resolution, and localizes specific knowledge gaps. It significantly enhances personalized instructional support while providing interpretable, pedagogically grounded insights—offering a transparent, equity-oriented technical pathway for adaptive education.

To address subjectivity, high labor costs, and insufficient cultural adaptation in classroom observation, this study develops the first multimodal AI behavioral analysis system tailored for Asian classrooms. Methodologically, it integrates audio, video, and environmental sensor data; proposes a culturally adaptive educational AI analytics framework; designs a scoring-free, feedback-oriented instructional reflection dashboard; and establishes the first publicly available audiovisual annotation dataset of Asian classroom interactions. Key contributions include: (1) introducing interpretable, behavior-based classroom metrics; (2) achieving real-time behavioral recognition with low cognitive load and high usability—validated by eight experts from Singapore’s National Institute of Education; and (3) significantly reducing human effort required for classroom observation. The system provides teachers with objective, scalable, and context-sensitive technological support for professional development.

This study addresses the tension between exploratory freedom and cognitive load in formative feedback by proposing an error-specific, dynamically calibrated automated feedback strategy for linear and exponential extrapolation tasks—balancing student autonomy with timely scaffolding. Employing a mixed-methods approach—including screen recordings, learning platform interaction logs, and structured interviews—we empirically examined student interaction patterns and feedback receptivity. Results indicate that the strategy significantly increases students’ self-initiated error correction (p < 0.01); 87% of participants rated its prompting intensity as “just right.” Compared to conventional immediate error correction, it better sustains engagement and reduces frustration. Its core contribution lies in being the first to embed an error typology specific to extrapolation tasks into the feedback triggering mechanism, enabling real-time alignment of support intensity with evolving cognitive demands. This yields a reusable theoretical framework and empirical foundation for adaptive, human–AI collaborative instructional feedback design.

This paper addresses the distortion of effect size estimates in educational and psychological intervention research due to differential item functioning (DIF). Moving beyond conventional differential test functioning (DTF) analyses that rely on total-score differences, we propose a novel causal robustness framework grounded in item response theory (IRT). We formally define “impact” as between-group differences in the latent trait distribution and develop a Hausman-type test that integrates DIF modeling directly into causal effect identification—thereby disentangling true construct-level impact from item-specific bias. Methodologically, we introduce a DIF-robust doubly robust estimator and a testable framework for effect generalizability inference. Empirical validation across item-level data from 34 randomized trials shows that DIF correction substantially reduces discrepancies between effect estimates derived from researcher-developed versus independent measures, thereby enhancing construct validity and cross-measure comparability of effect interpretations.

This study addresses the challenge of uncovering deep-seated misconceptions among students regarding core difficult topics in online medical courses. To this end, it proposes a two-stage, multi-source data analysis approach that integrates quiz performance, response behaviors, and lecture transcripts. The first stage identifies high-challenge topics, while the second leverages large language models to detect cognitive biases that conventional metrics often miss. Findings are validated through expert instructor interviews and quantitative indicators. This work represents the first systematic integration of behavioral data with course content to reveal implicit misunderstandings using large language models, earning strong endorsement from domain experts. The approach successfully pinpoints several critical misconceptions, offering a scalable and practical pathway for targeted instructional interventions and efficacy evaluation.

This study investigates how students’ background advantages influence learning outcomes through their interaction behaviors with generative AI. Conducting a controlled experiment with 318 undergraduate students and integrating behavioral coding with statistical modeling, the research systematically uncovers—for the first time—the mediating role of learning behaviors between background characteristics and academic performance. Findings reveal that students with higher background advantage tend to adopt proactive and critical interaction strategies, which significantly enhance learning effectiveness. Notably, once behavioral variables are accounted for, the direct effect of background factors on achievement substantially diminishes or disappears entirely, suggesting that optimizing interaction behaviors with generative AI can effectively mitigate disparities in learning outcomes across student groups.

This study investigates how micro-level feedback features in AI tutoring—such as elaboration specificity, empathetic language, and response length—affect students’ immediate comprehension and confusion states during sequential dialogues. Analyzing 1,718 naturally occurring interaction sequences from the StudyChat dataset, the authors employ chi-square tests and generalized estimating equations (GEE) to model these effects. The findings reveal that specific elaborations significantly enhance subsequent student understanding and reduce re-confusion, demonstrating a positive learning impact. In contrast, longer responses are associated with decreased comprehension, while empathetic language shows no statistically significant effect. This work is the first to uncover the differential roles of micro-feedback within authentic, continuous tutoring dialogues, challenging conventional assumptions regarding the benefits of extended replies and empathetic expressions in educational AI systems.

This study investigates whether student behaviors and academic performance in e-learning environments can reflect symptoms of depression and anxiety. Leveraging Moodle log data, course grades, and self-reported psychological scales, the research employs event log mining, descriptive statistics, Benjamini–Hochberg–corrected t-tests, effect size estimation, and Spearman correlation analyses to systematically identify significant associations between multidimensional learning management system (LMS) behavioral indicators and mental health symptoms. The findings reveal that depression is significantly associated with temporal shifts in activity patterns, increased session duration, and last-minute assignment submissions, whereas anxiety correlates strongly with the concentration of study time and variability in session characteristics. These results highlight the potential of such non-intrusive behavioral markers to serve as early warning signals for mental health concerns in educational settings.

This study addresses a critical gap in the evaluation of AI tutoring systems, which has traditionally emphasized the instructional quality of feedback while overlooking how students actually engage with and utilize it. The authors propose a novel assessment framework that integrates student behavioral data, introducing interaction-based dimensions to examine whether learners adopt and correctly apply AI-generated feedback. Through large-scale analysis of code submissions and interaction logs, the research demonstrates that behavioral signals are more effective than conventional instructional quality metrics in predicting students’ perceived usefulness of feedback. The framework’s validity is further confirmed across two consecutive semesters in authentic classroom settings, offering a new paradigm for the comprehensive evaluation of AI tutoring systems.