How to Boost the Performance of Recommender Systems by Social Trust? Studying the Challenges and Proposing a Solution

Abstract

With the increasing number of items in electronic retailers, news websites, etc., finding interesting items concerning the taste of users is becoming more challenging. Recommender Systems (RS) are a well-known solution to this issue. Collaborative filtering (CF) is a widely accepted and popular technique to implement an RS. However, cold-start and data sparsity problems reduce the performance of CF methods. One promising solution for these issues is to use the social trust information. However, how to properly use social trust information is a hot and still open question. In this paper, we propose a similarity measure and a simple link prediction method to address this question and employ them in trust-aware matrix factorization. Especially, our proposed similarity measure is asymmetric to consider the nature of social relationships. Also, to have a more accurate similarity estimation, we have considered both the user's historical ratings and trust relations, and we have determined the weight of each source. Finally, we have used the item-based model and the level of interest a user's trustee has for an item to improve the performance of the proposed method for sparse datasets. We conduct extensive performance evaluations in terms of rate prediction and interesting items found. Experimental results on three real-world datasets demonstrate the effectiveness of the proposed method, especially in terms of Mean Absolute Error.