The Algorithm of Sequential Music Recommendation Based on BERT

Music recommendation algorithms aim to predict the songs that users might be interested in based on various factors, such as their listening history. The best existing method is Bert_Music, a sequential music recommendation model based on Bidirectional Encoder Representation from Transformer (BERT). This model leverages user sequence context information to improve recommendation performance. However, Bert_Music only considers the sequence information of songs and ignores their content characteristics. To address this limitation, we propose an improved version of Bert_Music, which is a sequential music recommendation model based on Bidirectional Encoder Representation from Transformer with tag information (BERT_MUSIC_T), uses song ID (Identity) and tag information to make more accurate recommendations. We compared the performance of Bert_Music_T with that of the Bert_Music model on the Last.fm dataset and found that the Precision has increased by up to 2.85%. These results prove the validity of our method.