Improving Recommendation Accuracy by Combining Trust Communities and Collaborative Filtering
Xiao Ma, Hongwei Lu, Zaobin Gan · 2014
With the booming of online social networks, social trust has been used to cluster users in recommender systems. It has been proven to improve the recommendation accuracy when trust communities are integrated into memory-based collaborative filtering algorithms. However, existing trust community mining methods only consider the trust relationships, regardless of the distrust information. In this paper, considering both the trust and distrust relationships, a SVD signs based community mining method is proposed to process the trust relationship matrix in order to discover the trust communities. A modified trust metric which considers a given user's expertise level in a community is presented to obtain the indirect trust values between users. Then some missing ratings of the given user are complemented by the weighted average preference of his/her trusted neighbors selected in the same community during the random walk procedures. Finally, the prediction for a given item is generated by the conventional collaborative filtering. The comparison experiments on Epinions data set demonstrate that our approach outperforms other state-of-the-art methods in terms of RMSE and RC.