Abstract: The traditional item-based collaborative filtering algorithm only focuses on the rating data without the chara-cteristics of items when calculating the similarity between items.The appearance of social tagging can reflect the characteristics of items,but there are some semantic fuzziness problems while adding the social tags into the collaborative filtering algorithm directly.To solve the problems above,this paper put forward an improved item-based collaborative filtering recommendation algorithm.It clusters social tags to generate tag clusters which represent different topics,and calculates the relevance between items and topics to generate item-topics matrix according to the tagging results of items.The similarity between items is calculated by combining item-topics matrix with item-ratings matrix,the rating of target items are predicted through the collaborative filtering algorithm,and the personalized recommendation is realized.Expe-rimental results on MovieLens dataset show that the proposed algorithm can eliminate the semantic fuzziness and improve the quality of recommendation.

Key words: Social tagging,Tag clustering,Item topics,Collaborative filtering,Personalized recommendation