Music Recommendation Algorithm Based on Dual-Stream Sequence Fusion

With the rapid development of computer technology and online service platforms, more and more music recommendation algorithms appear to recommend music that users may like. Most of the existing music recommendation algorithms use traditional machine learning algorithms based on content-based recommendation filtering, K-nearest neighbour classification algorithms, collaborative filtering, etc. However, such recommendation algorithms often suffer from poor scalability, and do not take into account the temporal nature of the music sequences as well as the interactions between the user and the songs, which results in poor recommendation results. In this paper, we propose a deep learning-based dual-stream music recommendation algorithm that combines a long and short-term memory neural network and an attention mechanism to capture features of audio data and predict users’ preferences for music by analyzing their historical preferences. We validate the performance of our proposed dual-stream music recommendation model on a million music dataset and compare it with existing sequence processing models, and the experiments show that the recommendation performance of our proposed algorithm outperforms that of existing methods.