Modeling Student Knowledge Progression Across Concepts in Intelligent Tutoring Interactions

Kanav Mittal, Abigail O'Neill, Hanna Schlegel, Gireeja Ranade, Narges Norouzi

Abstract: As Intelligent Tutoring Systems (ITS) are deployed at scale, developing detailed learner models is essential for validating their effectiveness and enabling iterative improvements, including personalized adaptation. We introduce a data-driven approach to modeling student knowledge progression across concepts in intelligent tutoring interactions. We prompt a Large Language Model (LLM) to determine which Knowledge Components (KCs) present in the exercise are used correctly in each student’s code submission. We then construct KC trajectories to analyze how students’ average KC understanding evolves in each exercise and across the semester. To predict both current and future KC performance for a specific student, we train a Long Short-Term Memory (LSTM) network that incorporates intelligent tutoring feedback and a student’s programming attempt history to predict performance on their next attempt of an exercise. Evaluating our framework on 8,000 student programming attempts demonstrates that our framework extracts meaningful insights about KCs learned in the course and outperforms baselines in predicting future KC performance.

External IDs:dblp:conf/aied/MittalOSRN25