Content and Popularity-Based Music Recommendation System

Abstract

The future of many modern technologies includes machine learning and deep learning methodologies. One of the prominent applications of these technologies is the recommender system. Due to the rapid growth of the songs in digital formats, the searching and managing of songs has become a great problem. In this study, the authors developed a recommender system using popularity and rhythm content of the song. The studies compared various techniques to improve the robustness and minimal error of the system. The authors will mostly focus on content-based, popularity-based, and collaborative-based filtering algorithms and also try to combine them using a hybrid approach. The authors utilized MAE for comparing the several procedures implemented here for the recommendation. Out of all procedures used, SVD performed well with MAE of 1.60 while KNN didn't perform that well as the authors had fewer features of song with mean absolute error of 2.212. User-relied and item-relied prototypes performed the best with MAE of 0.931 and 0.629.