A Collaborative Filtering Method using Topological-Potential Based Community Discovery Strategy

摘要

Collaborative filtering is one of the most successful technologies for recommender systems. However, it is strongly limited by the sparsity of data. To deal with these limitations, this paper presents an efficient community collaborative filtering method based on community structures. We apply multi-relational data mining techniques to construct a user network, which not only are the user-item ratings utilized but also user and item information, and then adopt a community discovery method based on topological potential to explore community structures, which in turn are used in collaborative filtering. We explore the optimal threshold of relational distance, and compare our method with other collaborative filtering methods. Experimental results show that the proposed community-based method gets lower MAE values and outperforms other methods. And we also find that increasing the size of training users and active users' rated items can effectively improve the prediction accuracy.