A collaborative filtering-based algorithm for music platform recommendations

The importance of recommendation algorithms in music platforms cannot be overstated. These algorithms are the backbone of personalized music experiences, enabling platforms to cater to the diverse tastes and preferences of millions of users worldwide. This study innovatively combines Graph Attention Networks (GATs) with collaborative filtering to enhance personalized recommendation systems. We created a user-song interaction matrix and utilized cosine similarity to establish user-based and song-based collaborative filtering models. Our experiments compared different weight allocation strategies and the impact of integrating GATs, revealing that the song-based approach outperformed across various datasets. The integration of GATs significantly improved accuracy, with our system achieving a 1.1% higher hit rate compared to traditional methods, highlighting the model's effectiveness. Additionally, the incorporation of random edge dropping and weighted initialization further optimized system performance, demonstrating the robustness of our approach in the realm of personalized recommendations.