A Novel Recommendation Model Based on Trust Relations and Item Ratings in Social Networks
Haokai Song, Qingqi Pei, Yang Xiao, Zi Li, Yong Wang · 2017
Recommender systems play an essential role in providing users with accurate and positive items or services for their personalized preferences from large volume of information choices. Collaborative filtering (CF) is an indispensable technique of recommender systems and widely applied in many areas such as e-commerce, social medium and review sites. However, CF suffers from three issues which are cold start users, cold start items and data sparsity. These issues severely degrade the recommendation performance of CF. To address these issues, we propose TrustTR, a novel memory-based recommendation model based on trust relations and item ratings. TrustTR integrates trusted friends' recommendations, item reputation and user history ratings to improve the recommendation performance. Especially, we calculate the user reputation based on trust relations and ratings information to supply the insufficient item ratings. Thus, TrustTR can alleviate the data sparsity problem and improve the quality of recommendations for cold start users. Besides, we utilize item reputation as one of the recommendation factors to mitigate the problem of cold start items. Finally, the experimental results from a real-world dataset show that our method can produce better predictive accuracy compared with other five counterparts recommendation models.