Abstract:

Recommender systems suggest a few items to the users by understanding their past behaviors. However, the existing collaborative filtering (CF) based recommender systems do not employ the information about latent item preference. In this article, a new CF personalized recommendation approaches was proposed. This approach aims to find user clusters using K-means clustering, and utilizes user clusters and utility matrix to construct item preference matrix,then, combines the item rating similarity, the item attribute and its preference features similarity in the item based CF process to produce recommendations. Experiments show the approach achieves the better result, but also to some extent alleviate the sparsity issue in the recommender systems.

Key words: collaborative filtering, recommender systems, personalized recommendation, data mining