A Music Recommendation System based on Reinforcement Learning

Abstract

At present, collaborative filtering recommendation method is the most widely used method. However, there are still two key problems in recommender system: cold start, exploration and exploitation. This paper uses multi-armed bandits (ϵ-greedy) algorithm in reinforcement learning to solve the problem of cold start and exploration-exploitation trade-off problem in music recommendation system. Experimental results show that compared with the traditional music recommendation method based on singular value decomposition, the music recommendation system based on this method can better meet the personalized needs of users.