Using user similarity to infer trust values in social networks regardless of direct ratings

Social networks recently get more attention on the Internet. Although they were introduced to facilitate relationships, now users may utilize them to get services such as experts' consultancy and marketing. Anyhow, users need a somehow proper estimation of trust in other users to make better decisions. Some trust evaluation mechanisms, which use direct ratings to calculate or propagate trust values, have been offered. However in social networks in which users only have binary relationship with each other, there is no direct rating value. Therefore a method is required to infer the values of trust and user reputation in social networks. In this paper, we propose a new method that employs user similarities to extract trust values without any need of direct rating. In our approach, user similarity is calculated from profile information and shared text via text-mining techniques. To show the effectiveness of our approach, it has been evaluated through rates gathered directly from the users. Comparing these rates with experimental results shows that the estimated trust values, obtained by this approach, are sufficiently acceptable. Besides the application of this approach in social networks, the proposed technique also can be used in direct rating mechanisms to evaluate correctness of trust values assigned by users, and consequently increase reliability of trust and reputation mechanisms against possible security threats.