Analysis of robustness in trust-based recommender systems

Much research has recently been carried out on the incorporation of trust models into recommender systems. It is generally understood that trust-based recommender systems can help to improve the accuracy of predictions. Moreover they provide greater robustness against profile injection attacks by malicious users. In this paper we analyze these contentions in the context of two trust-based algorithms. We note that one of the characteristics of trust-based algorithms is that ratings are often exposed in the user population in order for users to develop opinions on the trustworthiness of their peers. We will argue that exposing ratings presents a robustness vulnerability in these systems and we will show how this vulnerability can be exploited in the development of profile injection attacks. We conclude that the improved accuracy obtained in trust-based systems may well come at a cost of decreased robustness. In the end, trust models should be selected very carefully when building trust-based collaborative filtering (CF) systems.