Author(s)

Yikan Wang, Chenwei Gong, Qiming Xu, Yingqiao Zheng

Corresponding Author

Yikan Wang

Abstract

The protection of user data privacy has become a key challenge in data-driven personalized recommendation systems. Traditional centralized recommendation methods often require uploading user data to a server for centralized training, which has a high risk of data leakage. For this reason, this paper designs a privacy-preserving personalized recommendation system based on federated learning. In this system, the user data is kept in the local device, and the distributed training of the model is realized by the federated learning algorithm, so as to complete the recommendation task under the premise of guaranteeing the user's privacy. This paper analyzes the privacy-preserving mechanism of federated learning and proposes a system architecture design for personalized recommendation; the performance of the system is verified based on different experimental scenarios, and the accuracy, response speed and privacy-preserving effect of the model are evaluated. The experimental results show that the system effectively reduces the risk of user data exposure while improving the recommendation effect, providing a new solution for building a safe and trustworthy personalized recommendation system.

Keywords

federated learning, privacy protection, personalized recommender systems, distributed training, data security