Multiple Random Walks For Personalized Ranking With Trust And Distrust

Social networks with trust and distrust relationships has been an emerging topic, aiming at identifying users' friends and foes when sharing information in social networks or purchasing products online. In this study we investigate how to generate accurate personalized rankings while considering both trust and distrust user relationships. This paper includes the following contributions, first we propose a social inference step of missing (indirect) trust relationships via multiple random walks, while considering users' direct trust and distrust relationships during the inference. In doing so, we can better capture the missing trust relationships between users in an enhanced signed network. Then, we introduce a regularization framework to account for (i) the structural properties of the enhanced graph with the inferred trust relationships, and (ii) the user's trust and distrust personalized preferences in the graph to produce his/her personalized ranking list. We evaluate the performance of the proposed approach on a benchmark dataset from Slashdot. Our experiments demonstrate the superiority of the proposed approach over state-of-theart methods that also consider trust and distrust relationships in the personalized ranking task.