Differential Privacy in Recommender Systems: Balancing Personalization and Protection
Keywords:
Differential Privacy, Recommender Systems, Privacy-Utility Trade-off, Matrix Factorization, DP-SGD, Federated Learning, Membership Inference Attacks, Privacy BudgetAbstract
Recommender systems (RS) are critical for personalized user experiences but pose significantprivacy risks through inference attacks on user data. Differential privacy (DP) providesmathematically rigorous privacy guarantees by introducing calibrated noise into computations.This paper comprehensively analyzes DP mechanisms
References
Asad, M., Shaukat, S., Javanmardi, E., Nakazato, J., & Tsukada, M. (2023). A comprehensive survey on privacy-preserving techniques in federated recommendation systems. Applied Sciences, 13(10), 6201. https://doi.org/10.3390/app13106201
Hao, W., Mehta, N., Liang, K. J., Cheng, P., El-Khamy, M., & Carin, L. (2022). Waffle: Weight anonymized factorization for federated learning. IEEE Access, 10, 49207–49218. https://doi.org/10.1109/ACCESS.2022.3172945
