Abstract: Traditional collaborative filtering algorithm reflects the user interest preferences and similarity of items by user ratings．It ignores the characteristics of user and project,and performs not very well for sparse data and new items．Under the age of Web2.0,social tab allows the user to label resources based on personal preferences freedom．To solve the problems,a hybrid algorithm based on tags and collaborative filtering recommendation algorithm was proposed．The method uses the label as the user interest information and the item characteristic．Through making use of the multidimensional relationship of the user,social and labeling,algorithm generates user feature vector and Item feature,and calculates the user preferences for items and projects similarity．Then based on the historical behavior of the user,user preference on other projects is predicted．Finally,sorting the predicted preference,recommended results are generated．Experimental results show that our algorithm can effectively alleviate data sparsity,solve the cold start,and enhance the accuracy of the recommendation.

Key words: Tag,Collaborative filtering,Recommendation algorithm,User preference,Item similarity