In recent years, there has been growing focus on the study of automatedrecommender systems. Music recommendation systems serve as a prominent domainfor such works, both from an academic and a commercial perspective. Afundamental aspect of music perception is that music is experienced in temporalcontext and in sequence. In this work we present DJ-MC, a novelreinforcement-learning framework for music recommendation that does notrecommend songs individually but rather song sequences, or playlists, based ona model of preferences for both songs and song transitions. The model islearned online and is uniquely adapted for each listener. To reduce explorationtime, DJ-MC exploits user feedback to initialize a model, which it subsequentlyupdates by reinforcement. We evaluate our framework with human participantsusing both real song and playlist data. Our results indicate that DJ-MC'sability to recommend sequences of songs provides a significant improvement overmore straightforward approaches, which do not take transitions into account.
展开▼
