A Personalized Music Recommender Based on Potential Preference Learning Dynamically
An intelligent musical recommendation system for multi-users in network context is presented. The system is based on a comprehensive user profile described by feature-weight-like_degree-scene vectors. According different scenes, the system can filter the music that user may like in the internet, and form a music recommendation list which will be sent to the user. The Preference Learning Agent updates the users’ profile dynamically based on explicit feedback or the hidden preference obtained from the users’ behavior. The learning rate of like_degree, original like_degree and the weight of feature type are important for the improvement of the feature’s learning efficiency. The recommendation system can capture the users’ potential interest and the evolvement of preferences. Experiment results show that the algorithm can learn users’ preferences effectively.
Elwin Mao and Linli Xu
X. W. Shi et al., "A Personalized Music Recommender Based on Potential Preference Learning Dynamically", Advanced Engineering Forum, Vol. 1, pp. 395-399, 2011