Publication
Clickstream data from digital learning environments provide valuable insights into student behavior but are challenging to interpret due to their granularity. Prior methods mainly relied on handcrafted features, expert labeling, clustering, or supervised models, limiting generalizability and scalability. We present ClickSight, an in-context Large Language Model (LLM)-based pipeline that interprets student clickstreams given a list of learning strategies to generate textual interpretations of students’ behaviors during interaction. We evaluate four prompting strategies and assess the effect of self-refinement across two open-ended environments using domain-expert rubric-based evaluations. Results show that while LLMs can reasonably interpret learning strategies from clickstreams, interpretation quality varies by prompting strategy, and self-refinement offers limited improvement. ClickSight demonstrates the potential of LLMs to generate theory-driven insights from educational interaction data.