Trusting the Trusted: An Empirical Analysis

Abstract

One aspect of research in the area of bootstrapping of cold start users in trust-aware recommender systems is on guiding new users find trusted users in the trust network. In this paper we examine the question �Which type of users do cold start users actually trust?� We analyze the users trusted by cold start users in the Epinions dataset (Massa & Avesani, 2006a). We examine the set of trusted users on the basis of four critical parameters: number of outgoing links, number of incoming links, number of items rated and a hybrid parameter defined as preference score. We also experimentally evaluate the impact of the four parameters on prediction accuracy. Our analysis shows that cold start users will more likely trust a user with higher number of incoming links as a trusted user even though trusting users with high preference score would result in better prediction accuracy.