An Improved Collaborative Filtering Algorithm Based on Tags and User Ratings

Aiming at the problem that the existing social tags recommendation system in building user interest model does not fully reflect the genuine interests, this paper proposes aim proved recommended algorithm (TARBCF) based on tags and user ratings. Since the rating data often sparse, to make the best use of the both advantages of ratings and tags, a rating predicts algorithm based on item category is introduced to predict the ratings. In this paper, user’s ratings can be incorporated to calculate the weight of tags. Considering the user interest has the time characteristic, time window is used to capture the current interests of user. Thus, by analyzing the traditional collaborative filtering thought, considering the relationship between user ratings and tags as well as the influence of user’s current interest, this paper set up an user-tag correlation matrix, which can calculate the target user’s nearest neighbors. Then according to the neighbor users predict the target user’s preferences of candidate items. Finally, taking the top-N scores items recommend to the target user. Simulation experimental results show that the improved algorithm can better reflect the user's preferences, and the quality of its recommendations were superior to the traditional scheme.