Sirui Ren
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Sirui Ren has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Ha Nguyen
University of North Carolina at Chapel Hill
2 shared publications - Jeffrey Alan Greene
University of North Carolina at Chapel Hill, Chapel Hill, USA
2 shared publications - Linyu Yu
The University of North Carolina at Chapel Hill
2 shared publications - Matthew L. Bernacki
University of North Carolina at Chapel Hill, Chapel Hill, USA
2 shared publications Benjamin Kelcey
University of Cincinnati, Cincinnati, OH, USA
1 shared publication- Jessaca Spybrook
Western Michigan University, Kalamazoo, MI, USA
1 shared publication
Latest Publications
- Open Access
- 16 August 2026
- Published by Center for Open Science in PsyArXiv
Self-regulation shapes who benefits from large language models
Large language models (LLMs) promise to broaden access to expertise, yet their benefits are uneven: the same model can scaffold a novice, accelerate an expert, or produce polished work that masks shallow understanding. We apply the established framework of self-regulated learning to human–LLM interaction and argue that loop quality emerges from the user–task–interface system rather than residing in the user alone. Productive loops involve three observable moves: steering sets goals and standards; monitoring evaluates fluent but fallible output; and repair tests, critiques, and revises it. Weak loops can create a synthesis trap in which users offload comparison and integration, producing false mastery—better assisted performance without corresponding transfer, confidence calibration, or error detection—and accumulating verification debt. The framework predicts that LLMs will widen differences between users when verification burden is high and regulatory support is weak, but narrow them when tasks are verifiable and interfaces externalize standards, evidence, and checks. We propose constraint density, monitoring calibration, and repair depth as process markers linking interaction traces to independent outcomes. This account turns divergent findings into testable predictions and motivates cognitive workbenches designed to preserve verification, agency, and transfer.
- 16 August 2026
- Published by Center for Open Science in PsyArXiv
Open AccessSelf-regulation shapes who benefits from large language models
Large language models (LLMs) promise to broaden access to expertise, yet their benefits are uneven: the same model can scaffold a novice, accelerate an expert, or produce polished work that masks shallow understanding. We apply the established framework of self-regulated learning to human–LLM interaction and argue that loop quality emerges from the user–task–interface system rather than residing in the user alone. Productive loops involve three observable moves: steering sets goals and standards; monitoring evaluates fluent but fallible output; and repair tests, critiques, and revises it. Weak loops can create a synthesis trap in which users offload comparison and integration, producing false mastery—better assisted performance without corresponding transfer, confidence calibration, or error detection—and accumulating verification debt. The framework predicts that LLMs will widen differences between users when verification burden is high and regulatory support is weak, but narrow them when tasks are verifiable and interfaces externalize standards, evidence, and checks. We propose constraint density, monitoring calibration, and repair depth as process markers linking interaction traces to independent outcomes. This account turns divergent findings into testable predictions and motivates cognitive workbenches designed to preserve verification, agency, and transfer.
- Open Access
- 25 June 2026
- Published by Society for Learning Analytics Research in Journal of Learning Analytics
- Vol. 13 (2) , 1-24
Using Large Language Models for Automated Coding of Self-Regulated Learning Think-Aloud Protocol Data
Documenting and understanding self-regulated learning (SRL) processes can inform the design of learning activities and scaffolds to enhance student success in STEM. Think-aloud protocols (TAPs)—prompting students to verbalize thoughts during task performance—reveal real-time, ecologically sensitive verbalizations of students’ SRL processes such as planning, monitoring, and strategy use. However, coding TAP data requires substantial resources. We investigated the capabilities of Large Language Models (LLMs) in automating the coding of SRL processes in TAPs across undergraduate STEM courses. To examine how task features, prompt engineering strategies, and SRL codes are linked to LLMs’ coding accuracy, we used a factorial design comparing different LLMs (GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro) across six SRL TAP codes (representing distinct cognitive and metacognitive processes), six prompt conditions (varying few-shots and context levels), and two STEM tasks (mathematics and biology). Analysis of 600 student verbalizations revealed that mathematics tasks yielded significantly higher classification accuracy compared to biology tasks, with GPT-4o and Claude 3.5 Sonnet outperforming Gemini 1.5 Pro. Few-shot prompting showed code-specific effects, with descriptively higher accuracy for monitoring negative judgments of learning but significantly decreased accuracy for subgoal setting. The addition of contextual information showed minimal impact across tasks. Cognitive process TAP codes (e.g., mathematical problem-solving) demonstrated the most consistent cross-task classification accuracy. In contrast, metacognitive monitoring (e.g., judgment of learning) showed substantial task-dependent variations. These findings highlighted both the promise and limitations of LLMs for scaling SRL research. They suggest that theoretical alignment in prompt engineering is essential for effective automated coding of dynamic regulatory processes.
- 25 June 2026
- Published by Society for Learning Analytics Research in Journal of Learning Analytics
- Vol. 13 (2) , 1-24
Open AccessUsing Large Language Models for Automated Coding of Self-Regulated Learning Think-Aloud Protocol Data
Documenting and understanding self-regulated learning (SRL) processes can inform the design of learning activities and scaffolds to enhance student success in STEM. Think-aloud protocols (TAPs)—prompting students to verbalize thoughts during task performance—reveal real-time, ecologically sensitive verbalizations of students’ SRL processes such as planning, monitoring, and strategy use. However, coding TAP data requires substantial resources. We investigated the capabilities of Large Language Models (LLMs) in automating the coding of SRL processes in TAPs across undergraduate STEM courses. To examine how task features, prompt engineering strategies, and SRL codes are linked to LLMs’ coding accuracy, we used a factorial design comparing different LLMs (GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro) across six SRL TAP codes (representing distinct cognitive and metacognitive processes), six prompt conditions (varying few-shots and context levels), and two STEM tasks (mathematics and biology). Analysis of 600 student verbalizations revealed that mathematics tasks yielded significantly higher classification accuracy compared to biology tasks, with GPT-4o and Claude 3.5 Sonnet outperforming Gemini 1.5 Pro. Few-shot prompting showed code-specific effects, with descriptively higher accuracy for monitoring negative judgments of learning but significantly decreased accuracy for subgoal setting. The addition of contextual information showed minimal impact across tasks. Cognitive process TAP codes (e.g., mathematical problem-solving) demonstrated the most consistent cross-task classification accuracy. In contrast, metacognitive monitoring (e.g., judgment of learning) showed substantial task-dependent variations. These findings highlighted both the promise and limitations of LLMs for scaling SRL research. They suggest that theoretical alignment in prompt engineering is essential for effective automated coding of dynamic regulatory processes.
- Open Access
- 26 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 786-792
Evaluating Large Language Model Feedback to Support Teacher Professional Vision: Prompting Strategies and Pre-service Teacher Engagement
- 26 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 786-792
Open AccessEvaluating Large Language Model Feedback to Support Teacher Professional Vision: Prompting Strategies and Pre-service Teacher Engagement
- Open Access
- 26 November 2025
- Published by Society for Learning Analytics Research in Journal of Learning Analytics
- Vol. 12 (3) , 66-86
Interpreting Predictive Learning Sequences in a College Math Course through a Self-Regulated Learning Framework
Digital traces have been used to measure self-regulated learning (SRL), yet the validity of inferences made about these traces has often been questioned. Recently, researchers have used multiple channels of data — including digital traces, verbalizations, and self-reports — to validate inferences about individual SRL events. Research on the validation of inferences about sequences of multiple SRL events remains limited; however, investigating these sequences has the potential to refine SRL theories. To study the validation of sequences of SRL events, we collected multimodal data from 49 undergraduates completing a math task in a lab setting. Participants were asked to think aloud while interacting with different digital platforms. Then, we used sequence pattern mining to identify the digital events most predictive of post-test scores. Next, we used student verbalizations during the learning process to validate the inferences about what those predictive sequences reflected. Sequences representing learner conscientiousness predicted better performance; sequences that included pausing and rewinding videos predicted poorer performance. Some learner verbalizations co-occurred with digital events and consistently aligned with SRL processes, providing validity evidence for SRL sequences. Heterogeneity in verbal-to-digital trace alignment emerged and will require methodological advances to validate the sequences specific to individuals and task conditions.
- 26 November 2025
- Published by Society for Learning Analytics Research in Journal of Learning Analytics
- Vol. 12 (3) , 66-86
Open AccessInterpreting Predictive Learning Sequences in a College Math Course through a Self-Regulated Learning Framework
Digital traces have been used to measure self-regulated learning (SRL), yet the validity of inferences made about these traces has often been questioned. Recently, researchers have used multiple channels of data — including digital traces, verbalizations, and self-reports — to validate inferences about individual SRL events. Research on the validation of inferences about sequences of multiple SRL events remains limited; however, investigating these sequences has the potential to refine SRL theories. To study the validation of sequences of SRL events, we collected multimodal data from 49 undergraduates completing a math task in a lab setting. Participants were asked to think aloud while interacting with different digital platforms. Then, we used sequence pattern mining to identify the digital events most predictive of post-test scores. Next, we used student verbalizations during the learning process to validate the inferences about what those predictive sequences reflected. Sequences representing learner conscientiousness predicted better performance; sequences that included pausing and rewinding videos predicted poorer performance. Some learner verbalizations co-occurred with digital events and consistently aligned with SRL processes, providing validity evidence for SRL sequences. Heterogeneity in verbal-to-digital trace alignment emerged and will require methodological advances to validate the sequences specific to individuals and task conditions.
- 25 July 2025
- Published by Elsevier in Journal of School Psychology
- Vol. 111, 101473
A practical guide to causal moderation analysis for investigating the role of context, identity, and culture in intervention research
- 25 July 2025
- Published by Elsevier in Journal of School Psychology
- Vol. 111, 101473
A practical guide to causal moderation analysis for investigating the role of context, identity, and culture in intervention research
- 8 March 2025
- Published by Springer Nature in Psychological Research
- Vol. 89 (2) , 1-23
Emotional states affect the degree of duration distortion more than distortion direction: a meta-analytic research
The impact of induced emotional states on duration perception remains unclear. This study investigated duration distortion in emotional states through two meta-analyses based on 17 English articles (28 studies) retrieved from the PubMed and Web of Science databases, published up to March 19, 2024. Meta-Analysis 1 examined both the distortion direction (i.e., underestimation or overestimation compared to neutral states) and the distortion degree (i.e., the extent to which perceived duration deviated from that in neutral states). Meta-Analysis 2 focused solely on the degree of distortion. The results indicated that when both the direction and degree were considered, the findings on duration distortion in emotional states were inconsistent. However, when the direction of distortion was not considered, there was a significant effect on the degree of duration distortion in emotional states. Moderator analysis revealed that the emotional valence moderated the effect size of emotion states on the degree of duration distortion, regardless of whether the direction of distortion was considered. While duration measurement paradigm only moderates the degree of duration distortion without direction. Specifically, the duration overestimation effect was observed in negative states when the distortion direction was considered. A greater distortion degree was observed in negative emotional states or duration judgment tasks. However, neither emotional arousal nor the duration itself moderated the distortion degree, irrespective of whether the direction of distortion was considered. These findings underscore the importance of considering the influence of induced emotional states on the degree of duration distortion, in addition to the direction.
- 8 March 2025
- Published by Springer Nature in Psychological Research
- Vol. 89 (2) , 1-23
Emotional states affect the degree of duration distortion more than distortion direction: a meta-analytic research
The impact of induced emotional states on duration perception remains unclear. This study investigated duration distortion in emotional states through two meta-analyses based on 17 English articles (28 studies) retrieved from the PubMed and Web of Science databases, published up to March 19, 2024. Meta-Analysis 1 examined both the distortion direction (i.e., underestimation or overestimation compared to neutral states) and the distortion degree (i.e., the extent to which perceived duration deviated from that in neutral states). Meta-Analysis 2 focused solely on the degree of distortion. The results indicated that when both the direction and degree were considered, the findings on duration distortion in emotional states were inconsistent. However, when the direction of distortion was not considered, there was a significant effect on the degree of duration distortion in emotional states. Moderator analysis revealed that the emotional valence moderated the effect size of emotion states on the degree of duration distortion, regardless of whether the direction of distortion was considered. While duration measurement paradigm only moderates the degree of duration distortion without direction. Specifically, the duration overestimation effect was observed in negative states when the distortion direction was considered. A greater distortion degree was observed in negative emotional states or duration judgment tasks. However, neither emotional arousal nor the duration itself moderated the distortion degree, irrespective of whether the direction of distortion was considered. These findings underscore the importance of considering the influence of induced emotional states on the degree of duration distortion, in addition to the direction.