Collaborative Filtering - Safe and Sound?

Recommender systems are now widely used to overcome the information overload problem. Security is an ever-present concern, and since many of these systems operate in an on-line environment, it is necessary to understand and remove any security issues that may exist. In earlier work, we have demonstrated that collaborative recommender systems are indeed vulnerable to biased noise attacks. In this paper, we expand on our previous work by adapting techniques from the field of Reputation Reporting Systems that substantially improve the robustness of recommender systems.