Differentially private neighborhood-based multi-criteria recommender system

Recommender systems are essential tools that provide personalized content suggestions by analyzing users' past preferences and interactions. Collaborative filtering techniques, widely used in these systems, generate predictions based on user similarities. In contrast to the limited structure of single-criteria systems, multi-criteria collaborative filtering approaches stand out with their potential to provide more accurate and personalized recommendations. However, the extensive use of user data in recommender systems brings significant privacy risks. While most studies in the literature assume that service providers are trustworthy, this assumption cannot always be guaranteed. Therefore, approaches based on local differential privacy, which protect data on the user side, have become increasingly important. In this study, a novel privacy-preserving multi-criteria collaborative filtering approach is proposed using three different local differential privacy -based differential privacy mechanisms: the Standard Laplace Mechanism, the Bounded Laplace Mechanism and the Truncated Laplace Mechanism. Experiments were conducted on two real-world datasets to evaluate recommendation accuracy under different privacy levels. The results show that the proposed method preserves privacy while maintaining recommendation quality. In this regard, the study offers a significant contribution to the integration of local differantial privacy into multi-criteria recommender systems. Keywords: Collaborative filtering, Differential privacy, Multi-criteria recommender system