A Content-Based Music Recommendation System Using RapidMiner

With the evolution of music information retrieval methods and music exploration technologies, there is enormous data generated on musical structure as well as users’ listening habits which can support new applications and innovations in this domain. This paper presents an overview of music recommender systems and explains the various state-of-the-art methodologies used to develop accurate music recommenders that regard both the objective and subjective effects of music on users. It also presents an application of music recommendation, particularly rating prediction and genre-based song recommendation which is implemented in RapidMiner. As a comparative study, we have analyzed user KNN, matrix factorization, and a combined model using user KNN and global average algorithms and evaluated their respective performances. We have found that hybrid recommender systems using combined models perform better than most single model variants. The song recommendation model is further demonstrated, and the ten most similar songs (based on the artist name specified by the user) are returned. Hence, our paper is focused on two central tasks—a brief review of existing methods as well the application of music recommender systems using the data science platform, RapidMiner which provides automation of machine learning tasks, portability, and functionality.