A Bayesian Multiagent Trust Model for Social Networks

Noel Sardana, Robin Cohen, Jie Zhang, Shuo Chen

Abstract: In this paper, we introduce a framework for modeling the trustworthiness of peers in the setting of online social networks. In these contexts, it may be important to be filtering the wealth of messages that have been sent, which form the ongoing communication within a large community of users. This is achieved by constructing an intelligent agent that reasons about the message and each peer rater of the message, learning over time to properly gage whether a message is good or bad to show a user, based on message ratings, rater similarity, and rater credibility. Our approach employs a partially observable Markov decision process for trust modeling, moving beyond the more traditional adoption of probabilistic reasoning using beta reputation functions. In addition to outlining the technique in full, we present empirical results to demonstrate the effectiveness of our methods, both in simulations featuring head to head comparisons with competitors, and in the context of some existing online social networks where ground truth data are available.