Large language models (LLMs) present new opportunities for creating pedagogical agents that engage in meaningful dialogue to support student learning. However, current LLM systems used in classrooms often lack the solid theoretical foundations found in earlier intelligent tutoring systems. To bridge this gap, we propose a framework that combines Evidence-Centered Design with Social Cognitive Theory and Zone of Proximal Development for adaptive scaffolding in LLM-based agents focused on STEM+C learning. We instantiate this framework with Inquizzitor, an LLM-based formative assessment agent that integrates human-AI hybrid intelligence and provides feedback grounded in cognitive science principles. Our findings show that Inquizzitor delivers high-quality assessment and interaction aligned with core learning theories, offering effective guidance that students value. This research demonstrates the potential for theory-driven LLM integration in education, highlighting the ability of these systems to provide adaptive and principled instruction.

This paper presents an approach that uses Large Language Models (LLMs) as late-fusion interpreters to synthesize multimodal signals from embodied classroom activities and infer students’ metacognitive behaviors. Our multimodal pipeline analyzes students’ movements, gaze, gestures, and speech within a mixed-reality simulation displayed on a classroom screen to support enactment and learning. Vision- and speech-derived features are fused at the interpretive layer via zero-shot prompting, self-consistency reasoning, and targeted prompt engineering to derive planning, enacting, monitoring, reflecting, and interacting behaviors. We investigate whether LLMs can reliably integrate modality-specific analytics to produce accurate behavioral labeling and whether an LLM-as-a-Judge can validate them at scale. To address scalability and reduce human burden, we introduce an automated evaluation protocol employing LLM-as-a-Judge to assess classification quality, enabling rapid, iterative benchmarking of model variants and prompt strategies. Using a balanced corpus of human-validated segments and perturbed controls, we compare text-only language models (e.g., GPT-5) with visual–language models (e.g., Qwen2.5-VL) that incorporate direct visual processing. Results indicate late-fusion, text-based LLMs can outperform VLMs on behavior judgment without raw video, and precision- or recall-oriented prompts adjust decision boundaries for subtle or brief segments. These findings position LLMs as effective late-fusion mechanisms for multimodal learning analytics and demonstrate the viability of LLM-as-a-Judge for scalable, human-in-the-loop evaluation.

Problem-based learning (PBL) environments increasingly embed LLM-based conversational agents (CAs) to scaffold inquiry, yet little is known about how learners actually respond to these agents in authentic classroom settings. Learning analytics offers powerful opportunities to capture and interpret how students engage with these agents, enabling deeper understanding of their inquiry processes and informing more adaptive instructional support in PBL settings. In this paper, we examine students’ interactions with three types of LLM-powered CAs — Content Knowledge, Argument Feedback, Argument Evaluation — designed to provide distinct forms of inquiry support within a narrative-centered learning environment. Using Pedaste et al.’s inquiry cycle as a lens, we used contextualized log data from 15 student groups to analyze how these agents shaped inquiry via sequence analysis of students’ coded actions. Our results revealed distinct trajectories of agent episodes and suggest LLM-powered CAs can play complementary pedagogical roles — supporting information seeking, guiding revision, and prompting reflection — but may also channel inquiry in ways that constrain exploration. We discuss the implications of using learning analytics to design adaptive scaffolds and using contextualized log analysis to capture how learners navigate inquiry with AI support in authentic classroom settings.

The field of learning analytics has made notable strides in automating the detection of complex learning processes in multimodal data. However, most advancements have focused on individualized problem-solving instead of collaborative, open-ended problem-solving, which may offer both affordances (richer data) and challenges (low cohesion) to behavioral prediction. Here, we extend predictive models to automatically detect socially shared regulation of learning (SSRL) behaviors in collaborative computational modeling environments using embedding-based approaches. We leverage large language models (LLMs) as summarization tools to generate task-aware representations of student dialogue aligned with system logs. These summaries, combined with text-only embeddings, context-enriched embeddings, and log-derived features, were used to train predictive models. Results show that text-only embeddings often achieve stronger performance in detecting SSRL behaviors related to enactment or group dynamics (e.g., off-task behavior or requesting assistance). In contrast, contextual and multimodal features provide complementary benefits for constructs such as planning and reflection. Overall, our findings highlight the promise of embedding-based models for extending learning analytics by enabling scalable detection of SSRL behaviors, ultimately supporting real-time feedback and adaptive scaffolding in collaborative learning environments that teachers value.

This paper explores the use of large language models (LLMs) to score and explain short-answer assessments in K-12 science. While existing methods can score more structured math and computer science assessments, they often do not provide explanations for the scores. Our study focuses on employing GPT-4 for automated assessment in middle school Earth Science, combining few-shot and active learning with chain-of-thought reasoning. Using a human-in-the-loop approach, we successfully score and provide meaningful explanations for formative assessment responses. A systematic analysis of our method's pros and cons sheds light on the potential for human-in-the-loop techniques to enhance automated grading for open-ended science assessments.

This study introduces a methodology to investigate students’ collaborative behaviors as they work in pairs to build computational models of scientific processes. We expand the Self-Regulated Learning (SRL) framework—specifically, Planning, Enacting, and Reflection—proposed in the literature, applying it to examine students’ collaborative problem-solving (CPS) behaviors in a computational modeling task. We analyze these behaviors by employing a Markov Chain (MC) modeling approach that scrutinizes students’ model construction and model debugging behaviors during CPS. This involves interpreting their actions in the system collected through computer logs and analyzing their conversations using a Large Language Model (LLM) as they progress through their modeling task in segments. Our analytical framework assesses the behaviors of high- and low-performing students by evaluating their proficiency in completing the specified computational model for a kinematics problem. We employ a mixed-methods approach, combining Markov Chain analysis of student problem-solving transitions with qualitative interpretations of their conversation segments. The results highlight distinct differences in behaviors between high- and low-performing groups, suggesting potential for developing adaptive scaffolds in future work to enhance support for students in collaborative problem-solving.

Eye-tracking technology has expanded our ability to quantitatively measure human perception. This rich data source has been widely used to characterize human behavior and cognition. However, eye-tracking analysis has been limited in its applicability, as contextualizing gaze to environmental artifacts is non-trivial. Moreover, the temporal evolution of gaze behavior through open-ended environments where learners are alternating between tasks often remains unclear. In this paper, we propose temporal fully-weighted scanpath graphs as a novel representation of gaze behavior and combine it with a clustering scheme to obtain high-level gaze summaries that can be mapped to cognitive tasks via network metrics and cluster mean graphs. In a case study with nurse simulation-based team training, our approach was able to explain changes in gaze behavior with respect to key events during the simulation. By identifying cognitive tasks via gaze behavior, learners’ strategies can be evaluated to create online performance metrics and personalized feedback.