Improving Recommendation Accuracy with Clustering-Based Social Regularization

Abstract

Social network information has been proven to be beneficial to improve the recommendation accuracy. Some recent works show a user may trust different subsets of friends regarding different domains concerning the heterogeneity and diversity of social relationships. However, these works obtain the friends subsets mainly by dividing friendships depending on the item categories, which aggravate the sparsity problem of social relationships and limit the contribution of social recommendation. In order to solve the issue, in this paper, we propose a novel social recommendation model by incorporating the friendships from different clustering-based user groups. We first formalize the user-preference matrix which describes the preferences of users from multiple domains and obtain the user groups by using the PDSFCM (Partial Distance Strategy Fuzzy C-Means) algorithm. Then we define the clustering-specific friends subsets and design a clustering-based social regularization term to integrate these friendships into the traditional Matrix Factorization model. The comparison experiments on Epinions data set demonstrate that our approach outperforms other state-of-the-art methods in terms of RMSE and MAE.