Modeling Trust for Rating Prediction in Recommender Systems

Traditional recommender systems usually ignore the social interactions between users in a social network and assume that users are independent and identically distributed. This assumption hinders the users to have access to personalized recommendations based on their circle of trusted friends. To model the recommender systems more accurately and realistically, we propose a social trust model and use the probabilistic matrix factorization method to predict user rating for products based on user-item rating matrix. The effect of users friends tastes is modeled using a real-valued trust which is defined based on importance and similarity between users. Similarity is modeled using a rating-based (Vector Space Similarity algorithm) and connection-based methods; centrality is quantified using degree and eigen-vector centralities. To validate the proposed method, rating estimation is performed on the Epinions dataset. Experiments show that our method provides better prediction when using trust relationship based on centrality and similarity rather than using the binary values. Also, degree centrality is shown to be more effective compared to the eigen-vector centrality. In addition, trust model using connection-based similarity is observed to have better performance compared to the ones that use rating-based similarity.