A Machine Learning Based Trust Evaluation Framework for Online Social Networks

Trust in Online Social Networks is an increasing popular issue in the field of social computing. It has been widely used in scenarios such as e-commerce, friend recommendation and trust-based access control system. Several trust computing methods have been proposed from different perspectives, however, most of them just quantify certain trust related factors and integrate them into a trust value by setting a weight for each factor. On the one hand, the weight is difficult to select since the influence that each factor has on trust is uncertain. On the other hand, the trust values generated by these methods might not be in accordance with users´ subjective judgements and they are not straightforward for users to understand. It is necessary for a trust evaluation method to provide accurate and intuitive trust levels for the social network users. Therefore, in this paper, trust evaluation is formalized as a classification problem and a novel approach utilizing machine learning method is presented. Firstly, the trust feature vector is constructed according to the trust related factors. Then by training with collected sample data which contains trust feature vectors and trust ratings, a trust classifier can be established. Based on this approach, a general trust evaluation framework is proposed for network-level trust evaluation. The experiments in real social networks verify the feasibility of our framework and show that the trained trust classifier has a relatively high predicting accuracy.