Deep Reinforcement Learning-based Music Recommendation with Knowledge Graph Using Acoustic Features

Abstract

In this study, we propose a new deep reinforcement learning-based music recommendation method with knowledge graphs. With the rapid development of Web services, music-related content posted on platforms, such as YouTube, is increasing dramatically. Conventional recommendation methods based on knowledge graphs have struggled with the coldstart problem caused by a lack of user preference information. The proposed method can solve this problem by introducing acoustic feature edges in the constructed knowledge graph. Furthermore, we realize efficient search using a deep reinforcement learning algorithm on a dense knowledge graph introducing acoustic feature-based edges. The proposed method can make appropriate recommendations even with a small amount of user preference information by learning the optimal action of the agent. We confirm the effectiveness of the proposed method by comparing our method with several conventional and state-of-the-art recommendation methods.