A Local Trust Inferring Algorithm based on Reinforcement Learning DoubleDQN in Online Social Networks

Abstract

The development of online social network has greatly promoted the interaction between users. Trust plays an important role in social activities of social network, which can effectively avoid the risk from unreliable users. In fact, most users have no direct interaction, i.e., there is no direct connection in the social network, so the indirect trust of the target user can only be evaluated by the friends who contact indirectly. The propagation and aggregation methods of trust affect the results of trust evaluation to a great extent. The existing aggregation methods generally have the problem of low prediction accuracy. In addition, how to find a reliable path to propagate trust is also a major challenge. This paper proposes a DoubleDQNTrust (DDQNTrust) algorithm based on reinforcement learning DoubleDQN to find reliable trust paths. Secondly, based on standard collaborative filtering and considering the similarity between users, a new aggregation method is proposed. The experimental results on Filmtrust online social network data set show that DDQNTrust algorithm can effectively find reliable trust paths, and can evaluate trust with high prediction accuracy.