Comparing Collaborative Filtering Methods Based on User-Topic Ratings

User based collaborative filtering (CF) has been suc-cessfully applied into recommender system for years. The main idea of user based CF is to discover communities of users sharing similar interests. However, existing user based CF methods may be inaccurate due to the problem of data sparsity. One possible way to improve it is to ap-pend new data sources into user based CF. Tags which are added and generated by users is one of the new sources. In order to utilize tags effectively, user-topic based CF is pro-posed to extract features behind tags, assign them to topics, and measure users ’ preferences on these topics. In this pa-per, we conduct comparisons between two user-topic based CF methods based on different tag-topic relations. Both methods calculate user-topic preferences according to rat-ings of items and topic weights. Experiments are conducted on the data set of MovieLens. The results show that user-topic based CF method is better than user based CF both in computational efficiency and recommendation effect. The effects are significant especially when each tag belongs to multiple topics