Privacy-Preserving Hierarchical Federated Recommendation Systems
Yucheng Chen, Chenyuan Feng, Daquan Feng
Abstract
With growing concern on data privacy, traditional Recommendation System (RS) raises the risk of privacy disclosure since it needs to collect a large amount of personal data. To tackle this problem, implementing RS in a federated learning (FL) manner is proposed as an efficient approach. Although various solutions have been proposed to improve privacy of federated RS models, related works ignore the communication efficiency. Moreover, most of related works merely consider one server to coordination with all users, which might not suitable for large-scale networks. To protect privacy and reduce communication overhead, we propose a privacy-preserving hierarchical federated collaborative filtering scheme for the RS. Finally, we provide the simulation results to evaluate our proposed scheme, which show that our scheme can maintain good recommendation accuracy, preserve data privacy and reduce communication overhead.