On the Effect of High Order Reputation Information on Trust Prediction in Wikipedia´s Vote Network

When a user (the truster) in a social network is trying to solve the problem of guessing whether he/she will trust or not another user (the trustee) when he/she has not direct experience of the trustee behavior, then the truster is facing a Trust prediction problem. In this paper we deal with this problem as a classification problem based on reputation features of the target trustee. Reputation refers to the opinion that a third person (a witness) may have about the target trustee. Second and higher order reputation information comes from witnesses which have no direct contact with the trustee. The differences in the spread of relationships among users produce variable size reputation feature vectors, according to the relationships of each user. We propose probabilistic descriptors of the reputation information in order to reduce the feature vectors to the same size. In this paper we explore the prediction of Trust training some classifiers with reputation features extracted from first and second order relationships. We conclude that second order reputation features do not improve classification significatively.