Graph-Enhanced Multimodal Knowledge Tracing with Reinforcement Learning for Explainable Personalized Music Learning Path Optimization

Abstract

Personalized learning in AI-driven music education still faces challenges such as the difficulty of integrating multimodal learning evidence, the complexity of modeling music skill dependencies, and the lack of long-term optimization of learning pathways. To address these issues, this paper proposes a framework for optimizing personalized music learning pathways driven by multimodal knowledge tracking and reinforcement learning. This method integrates textual descriptions, sheet music images, audio performances, learning behaviors, evaluation results, and cognitive parameters to construct a dynamic model of the learner’s music skill state; further introduces a Music Skill Knowledge Graph (MSKG) to characterize the prerequisite relationships and structural dependencies among pitch accuracy, rhythm, sight-reading, technical proficiency, musical expressiveness, and interpretive ability; based on this, it models learning path recommendations as a sequential decision-making problem, using reinforcement learning to comprehensively optimize learning gains, task difficulty matching, path coherence, learner engagement, and prerequisite constraints. Experimental analysis indicates that the proposed method enhances the ability to predict learner states, improves the interpretability of the music skill evolution process, and provides effective support for the generation of personalized music learning paths.