On Privacy-preserving Context-aware Recommender System

Privacy is an important issue in Context-aware recommender systems (CARSs). In this paper, we propose a privacy-preserving CARS in which a user can limit the contextual information submitted to the server without sacrificing a significant recommendation accuracy. Specifically, for users, we introduce a client-side algorithm that the user can employ to generalize its context to some extent, in order to protect her privacy. For the recommendation server, two server-side recommendation algorithms are proposed, which operate under the condition that only a generalized user context is given. The experimental results have shown that, using our approaches, the user context privacy can be achieved without a significant degradation of the recommendation accuracy.