Who Am I? History-Aware Profiles for Student Simulation in Tutoring Dialogues

2026-05-28

AI Summary

Existing student simulation methods struggle to effectively model students’ historical learning behaviors and knowledge states, resulting in generated dialogues that lack authenticity and coherence. To address this limitation, this work proposes a history-aware student simulation framework that constructs dynamic learner profiles grounded in historical question-answering interactions and dialogue records, which then condition the prediction of student utterances in tutoring conversations. The framework employs a two-stage architecture comprising a profile generator and a conditional student simulator, and for the first time integrates reinforcement learning to jointly optimize both components. Evaluated on a newly collected dataset from a real-world mathematics learning platform, the proposed approach significantly outperforms baseline methods, demonstrating the effectiveness and novelty of incorporating historical information, the learner profiling mechanism, and the reinforcement learning–based optimization strategy.

Abstract

A key part of developing large language model (LLM)-powered, automated tutoring tools is student simulation, i.e., using LLMs to role-play as students, which can facilitate tutor model evaluation and training. Existing work mostly focuses on within-dialogue simulation, which lacks context on student knowledge and behavior, partly due to not grounding in past student question-answering or dialogue interactions. In this work, we introduce the task of history-conditioned student simulation, where the goal is to accurately predict student dialogue turns by leveraging information in the student's learning history. We propose a two-component framework in which a profile generator summarizes a student's history and a simulator predicts student turns conditioned on the resulting profile. We train both components with reinforcement learning (RL), yielding profiles optimized for faithful student simulation. We evaluate our method and baselines on the first-of-its-kind real-world dataset of student dialogues and question responses that we collect from a math learning platform. Extensive experiments show that our method significantly outperforms baselines, and demonstrate the importance of history, profiles, and RL training.

Problem

• student simulation
• history-aware
• tutoring dialogues
• learning history
• profile generation

Innovation

• history-conditioned student simulation
• profile generation
• reinforcement learning
• LLM-based tutoring
• student modeling