ADAPTIVE EXERCISE RECOMMENDATION FOR MUSIC SKILL DEVELOPMENT USING MACHINE LEARNING

Abstract

The development of intelligent systems with the ability to create personalized learning paths is one of the most popular research trends in the area of educational technologies. In this paper, we present an adaptive exercise recommendation module which is embedded into the MusicEval intelligent music learning platform. The algorithm of our module is based on a hybrid recommendation strategy which involves content-based filtering and collaborative filtering to propose exercises tailored to a learner's performance gaps. We have created a database of 320 structured music exercises, classified according to three different skill types: pitch, rhythm, and dynamics. We have applied the recommendations generated by the algorithm to evaluate the performance of 4,200 audio samples. Our hybrid recommender system produced a Precision@5 score of 0.847 and an NDCG@10 of 0.881, which is significantly better than results obtained using a content-based and collaborative filtering algorithm independently.