Comparison between Collaborative Filtering and Neural Collaborative Filtering in Music Recommendation System

Abstract

Music is one of the most popular entertainments, and the music industry continues to increase over time. There are many types of genres in music, and everyone has their own choice of the type of music they want to listen to. The recommendation system is an important function in the application, especially when there are a large number of choices for a particular item. With a good recommendation system, users will be able to get help from the suggestions given and can improve the user experience of the application. By using collaborative filtering (CF) methods to recommend products related to personal preference history, this feature can be better provided. However, the CF method still lacks in integrating complex user data. Hybrid technology may be a solution to perfect the CF method. The combination of neural network and CF also called NCF is better than using CF alone. The focus of this research is a CF method combined with neural networks or neural collaborative filtering. In this study, we use 20,000 users, 6,000 songs, and 470,000 records of ratings then predict the score using CF and NCF approach. We aim to compare the recommendation systems using CF and NCF. The study shows that NCF is better in gathering certain playlists according to one's preferences, but it takes more time to build compared to user-based collaborative filtering.