Trust Inference in Online Social Networks

We study the problem of trust inference in signed social networks, in which, in addition to rating items, users can also indicate their disposition towards each other through directional signed links. We explore the problem in a semi-supervised setting, where given a small fraction of signed edges we classify the remaining edges by leveraging contextual information (i.e. the users' ratings). In order to model user behavior, we use deep learning algorithms i.e. a variation of Restricted Boltzmann machine and Autoencoders for user encoding and edge classification respectively. We evaluate our approach on a large-scale real-world dataset and show that it outperforms state-of-the art methods.