Attentive Context-Aware Music Recommendation

Listening to and singing music online have become two popular forms of entertainment for people. Online music service providers are devoted to recommend personalized music to users, so that more users can be attracted by spending more money on their music service platform. Hence, an accurate recommendation for users is one of the most important tasks for the music service companies. This paper proposes a novel recommender system for recommending music. According to the preliminary measurement, we find that the interest of users on music changes over time. Furthermore, we notice that the time interval between two consecutive listened/sung musics can reflect the temporal interests of users. Based on these findings, we propose MusicRecLSTM, which leverages the attention mechanism and a modified LSTM network to learn the embeddings of both the user and the music based on the sequential data and the temporal context. MusicRecLSTM also learns whether a user would tend to listen to a newly released music at a specific time. Evaluations over three real-world datasets show that the proposed MusicRecLSTM can provide high precision, compared to several state-of-the-art recommender systems, with at least 20% improvement on the recall and mean reciprocal metrics.