Trust prediction in multiplex networks

The proliferation of social networks and their popularity among web users has lead a lot of researches on their analysis. One of the social network classes is trust networks in which the links indicate trust or distrust. There are some challenges for these networks like interaction with anonymous users, inaccurate equations, time sensitivity and etc. Due to huge number of users, the first challenge has a high level of importance and one of its solutions is predicting trust and distrust values. In this manuscript we proposed a new method for trust and distrust prediction. To implement our method, first of all we constructed a multiplex network for our problem which consists of two layers and the relations in each layer have different concepts; one layer indicates trust relations and the other indicates similarity relations. We then ranked the nodes of this multiplex network using a degree ranking method specialized for this kind of networks and used these ranks to obtain optimism and reputation, which constructs our feature vectors for a logistic regression predictor. Our experiments on Epinions real data set, showed that accuracy of our proposed method in predicting trust and distrust relations is higher than the previous methods including the cluster-based collaborative filtering and some methods based on social theory, balance theory, and machine learning framework.