Demonstrating the value of credibility modeling for trust-based approaches to online message recommendation

This paper presents a validation of CredTrust, a multiagent trust model aimed at addressing the problem of folklore in online social networks: strong similarity among mis-guided peers should be ignored, at the expense of valuable advice provided by less similar credible peers. A series of simulations provides measured validation of the value of our approach in coping with a wide variety of messages, recommending those with high predicted benefit and avoiding those with low predicted benefit; we quantify our improvements in comparison with an established trust-based system (which fails to incorporate credibility).