DRL-MusicEdu: a deep reinforcement learning-based dynamic music teaching recommender system

Abstract

Addressing the inability of traditional music teaching systems to dynamically adapt to learners’ personalised states, this study proposes deep reinforcement learning-MusicEdu – a dynamic recommender system based on deep reinforcement learning. The framework constructs an intelligent agent that continuously perceives multidimensional learner states (skill proficiency, interests, fatigue) and dynamically optimises teaching-resource sequences via deep reinforcement learning (using proximal policy optimisation). This leverages a structured resource library derived from the Lakh Musical Instrument Digital Interface Dataset, annotated with metadata including difficulty, style, and technical attributes. Experimental validation across 20 weeks with five learner profiles demonstrates that deep reinforcement learning-MusicEdu significantly outperforms baselines, improving skill growth rate by 19.2% (p < 0.01) and user retention by 18.1%. The system enables personalised adaptive learning pathways, establishing an innovative decision-making framework for intelligent music education.