The success of a music recommendation (MR) system heavily relies on its ability to identify user needs. Existing approaches, including collaborative filtering and content- based methods, overlook the fact that user needs is inher- ently subjective and largely time-variant. In this work, we propose an interactive MR system (iMR) to tackle these issues. A user is asked to provide his/her preference for a number of songs, and then the feedback is exploited to learn the user needs. This way, the MR system is optimized for the user on the fly. To relieve user fatigue, the active interactive genetic algorithm is utilized in the learning pro- cess of user preference. In addition, to increase the hit rate, the songs awaiting for user evaluation are selected by the k-means algorithm. Experimental result demonstrates the efficacy and efficiency of the proposed system.

2009

English