Prototype for a Personalized Music Recommendation System Based on TL-CNN-GRU Model with 10-Fold Cross-Validation
Pei‐Chun Lin, Chen-Yu Yu, Eric Odle · 2024
Music is a medium that connects cultures, expresses emotions, and can create memory connections that are deeply embedded in people's hearts. With the rapid expansion of digital music platforms, providing an effective personalized recommendation system that helps users browse huge song libraries can effectively promote platform usage. However, existing music recommendation systems usually fail to fully capture individual preferences. In this study, we aim to address these limitations by developing an advanced recommendation system focusing on two facets: 1) building a model for music genre classification, and 2) providing a platform with personalized recommendations based on user preferences. We herein integrate three models for music genre classification, transfer learning, convolutional neural networks, and gated recurrent units (TL + CNN + GRU), using the GTZAN dataset. Results show that the TL + CNN + GRU model can improve the accuracy (55%->71%) of personalized music recommendation systems by using 10-fold cross-validation. Finally, we introduce a prototype platform for understanding user experience. In conclusion, our model not only improves the accuracy of recommendations, but also promotes user exposure to different music genres, redefining the user experience.