A new approach for rating-based collaborative music recommendation

In this paper, we propose a new rating-based collaborative music recommendation approach to help the users search their favorite music objects. This approach possesses the following characteristics, which overcomes the shortcomings existing in traditional rating-based collaborative recommendation systems. First, a semantic model of music based on perceptions is adopted for the users to rate their favorite music objects. To avoid using erroneous or useless information for collaborative recommendation, we identify the trustable users of each user group by applying data mining techniques on the associated rating histories. Only the ratings provided by the trustable users are considered for collaborative recommendation. Moreover, the personal preferences of the target user are also considered for recommendation. The personal preference is extracted from the rating history, which represents the perceptions of music objects the target user cares more. Finally, a feedback mechanism is provided to adjust the influence degrees of the personal preferences and the opinions of the corresponding trustable users for better recommendation. We implemented a music recommendation system based on this approach and performed experiments on this system to show the effectiveness of this approach.