Towards user-oriented privacy for recommender system data: A personalization-based approach to gender obfuscation for user profiles

In this paper, we propose a new privacy solution for the data used to train recommender system, i.e., user–item matrix. The matrix contains implicit information, which can be inferred using classifier, leading potential violations. Our solution, called Personalized Blurring (PerBlur), is simple, yet effective, approach adding and removing items from users’ profiles in order generate an obfuscated novelty of PerBlur personalization choice obfuscation individual user profiles. formulated within user-oriented paradigm system that aims at making solutions understandable, unobtrusive, useful user. When training, algorithm able reach performance comparable what attained when it trained on original, unobfuscated data. At same time, classifier no longer reliably use predict gender users, indicating information has been removed. addition introducing PerBlur, make several key contributions. First, evaluation protocol creates fair environment compare between different conditions. Second, carry out experiments show impacts fairness diversity results. sum, our work establishes transparent protect while time improving recommendation results users by maintaining enhancing diversity.