The Three-Tier Architecture of Federated Learning for Recommendation Systems

Abstract

Recommender systems are now vital in the Internet age to assist users in finding helpful stuff and reducing information overload. To assist users in finding personalized stuff, a large amount of user-sensitive data used for recommendations may lead to privacy violations. In recent research, federated learning-based recommender systems structures have made tremendous progress in boosting prediction accuracy while providing privacy. However, challenges still need to be concentrated on while employing federated learning 1) Ensuring user privacy and security of data and model privacy. 2) Heterogeneity of data in distributed entities to train a model with the best representation for better analysis, and 3) The communication between the user and server leads to increase overhead and latency. Developing a secured, privacy-protected recommender system that can accomplish high prediction accuracy is crucial and valuable. To address the above issues, a theoretical approach called a three-tier architectural solution is proposed to assure privacy guarantee without sacrificing accurate predictions on the recommendation with less overburden on a server. Further, discussed the future directions of recommendation systems by using federated learning.