Multi-Scale HyperGraph Node Embedding via Random Walking for Music Recommendation

Recommender systems are designed to deliver tailored recommendations to users, demonstrating efficacy in addressing the issue of data overload prevalent in numerous online platforms. Current music recommendation systems have produced considerable results, but due to the complexity and diversity of the interaction between users and music and the background of listening time, as well as the dynamics of user preferences, making full use of music listening records remains a challenging and important problem. Therefore, a novel Multi-Scale HyperGraph Node (MSHN) Embedding approach for sequential music recommendation is proposed in this paper. Firstly, a devised Weighted Random Walk Algorithm is used to learn the First-order Markov node sequence. Secondly, Users' preferences are subsequently inferred from their listening habits, leading to recommendations of music pieces that align with their immediate needs. Finally, extensive experiments are performed on two real-world datasets, demonstrating that MSHN outperforms baseline models.