A Federated Recommendation System Based on Local Differential Privacy Clustering

Traditional recommendation systems (RS) play an important role in applications such as electricity trade, e-commerce etc. However, there is a serious risk of data privacy leakage in traditional recommendation system (RS). To overcome the issues, Federated Learning (FL) and RS are employed for distributed training in recommendation system, which focuses on improving the accuracy to achieve similar performance as centralized training. However, it also causes bad training efficiency and privacy leakage. In this paper, we present a secure federated RS scheme based on local differential privacy and security aggregation to achieve centralized training-like performance while balancing data privacy protection and training efficiency. Finally, we demonstrate the performance of our scheme.