Privacy Risks for Multi-Criteria Collaborative Filtering Systems

Abstract

In case that individuals feel their privacy is violated while using any recommender system, they might be willing to declare incorrect information or even completely refuse to use such services. To relieve customer concerns, privacy risks that are inherent in the utilization of such systems need to be discussed principally, and service providers should offer privacy preservation mechanisms. Also, there shall be a balance between conflicting goals of accuracy and privacy. In the literature, researchers discuss privacy risks that users are exposed to due to the collection of personal preferences in collaborative recommender systems. However, such studies elaborate on threats arising by submitting a single preference value for items, and they fall short on evaluating privacy risks originating by the collection of preferences in a multi-criteria domain. It is an indisputable fact that accuracy of predictions is closely related to the quality of preference data. The collection of multi-criteria ratings provides a more fine-grained structure for creating dynamic user profiles and helps improve the quality of personalized recommendations. However, such multi-perspective preference data might confront users with more severe privacy problems. Therefore, motivating individuals toward the use of multi-criteria recommender systems rely on setting a balance between accuracy of provided predictions and ensured privacy levels. In this study, we evaluate existing privacy violation conditions from the perspective of multi-criteria recommender systems and discuss comprehensive threats exposed by such services.