查看文章

The study employed Large-Language Models (LLMs) to analyze students’ sequential patterns in self-regulated learning (SRL). A pre-trained LLM was fine-tuned on 106 high students’ computer trace data to learn the relationship between students’ SRL sequence and their final learning outcome. The LLM was validated against a linear baseline to demonstrate that LLM can model the SRL sequence. Furthermore, an integrated gradient method was employed to compute the sequence attribution to reveal how different sub-sequences contribute to the final learning outcome. The study found that the “evaluation-analysis” sequence had the highest attribution to a positive learning outcome. This finding demonstrated the importance of self-reflection in SRL. The proposed LLM-based analysis is possible to help educators design real-time interventions for different students to enhance their learning process.