Multi-faceted trust and distrust prediction for recommender systems

Many trust-aware recommender systems have explored the value of explicit trust, which is specified by users with binary values and simply treated as a concept with a single aspect. However, in social science, trust is known as a complex term with multiple facets, which have not been well exploited in prior recommender systems. In this paper, we attempt to address this issue by proposing a (dis)trust framework with considerations of both interpersonal and impersonal aspects of trust and distrust. Specifically, four interpersonal aspects (benevolence, competence, integrity and predictability) are computationally modelled based on users’ historic ratings, while impersonal aspects are formulated from the perspective of user connections in trust networks. Two logistic regression models are developed and trained by accommodating these factors, and then applied to predict continuous values of users’ trust and distrust, respectively. Trust information is further refined by corresponding predicted distrust information. The experimental results on real-world data sets demonstrate the effectiveness of our proposed model in further improving the performance of existing state-of-the-art trust-aware recommendation approaches.