A music recommendation system based on annotations about listeners' preferences and situations

In this paper, we present a playlist generation scheme that uses lyrics and annotations to discover similarity between kinds of music and user tastes. It generates a playlist according to user preferences and situations. Additionally, users can provide some feedbacks to the system such as whether each tune is suitable for the preference and the situation. The system transforms the feature values concerning preferences and situations and adapts them to each user. The playlists are generated through three phases. First, an initial playlist is found from databases by content-based retrieval. Second, transcoding improves the playlist according to the user's preference and situation. Finally, by interaction between the system and the user, the playlist becomes more suitable for the user.