INSIDE: A New Framework for Simulating Student Reasoning in LLMs

A new student modeling framework called INTERNAL STUDENT DIALOGUE (INSIDE) has been developed by researchers to enhance Large Language Models (LLMs) for producing internal dialogue aligned with Bloom's Taxonomy, addressing both cognitive and emotional aspects. This framework seeks to increase the accuracy of simulated student behaviors by incorporating the reasoning processes behind those actions. By fine-tuning models using paired think traces and actions, evaluations indicate notable improvements in the fidelity of actions and the quality of internal dialogue generated. This initiative fills a significant gap in LLM-based simulators, which typically replicate observable actions without addressing the reasoning behind them, a crucial factor in educational contexts like tutoring system assessments. The findings are published in a paper on arXiv (arXiv:2608.10492).