A community-based distrust-aware computational trust model for online social networks

Abstract

Abstract Trust calculation is an important aspect of social networks research which acts as the main pillar of the network’s existence. The research conducted in this domain still faces unsolved challenges that should be tackled to improve the accuracy of the results. One of these challenges is the cold start problem which we aim to reduce. We have achieved this goal by reducing the dependency of the trust calculation metrics on the links between two individual nodes. We calculate the trust score toward a node without the existence of a link between the trustor and a trustee according to individual characteristics and similarity factors. The consideration of both trust and distrust criteria in the model has resulted in higher accuracy and precision. The network is transformed into communities with nodes having similar subject interests by using topic modeling. The accuracy of the model presented in this research is not reduced despite the low number of communication edges. It has been shown in the evaluations that the mean absolute error of the trust calculations is about 0.01 even with the reduction of 90% of the edges. Hence, the model presented in this research is robust against the edge sparsity of the network. The proposed model is compared with notable current works namely the GCR method and the method presented by Akilal et al. Based on the average absolute error criterion, with a reduction of 90% of the edges, the accuracy of the proposed model shows a 98% increase compared to the GCR method and 96% increase compared to the Akilal et al.’s proposed method.