GEMRec: A Graph-Based Emotion-Aware Music Recommendation Approach

Music recommendation has gained substantial attention in recent times. As one of the most important context features, user emotion has great potential to improve recommendations, but this has not yet been sufficiently explored due to the difficulty of emotion acquisition and incorporation. This paper proposes a graph-based emotion-aware music recommendation approach (GEMRec) by simultaneously taking a user’s music listening history and emotion into consideration. The proposed approach models the relations between user, music, and emotion as a three-element tuple (user, music, emotion), upon which an Emotion Aware Graph (EAG) is built, and then a relevance propagation algorithm based on random walk is devised to rank the relevance of music items for recommendation. Evaluation experiments are conducted based on a real dataset collected from a Chinese microblog service in comparison to baselines. The results show that the emotional context from a user’s microblogs contributes to improving the performance of music recommendation in terms of hitrate, precision, recall, and F1 score.