A Review of Triangular Core Problem of Recommendation System Based on Federated Learning

Traditional recommender systems face severe challenges of data privacy protection. Centralized data storage and processing mode can not meet the compliance requirements of General Data Protection Regulation (GDPR), and there is a significant risk of privacy leakage. Federated learning provides a new solution for recommender systems through the distributed paradigm of "data does not move, model moves", and federated recommender systems emerge. It is found that the federal recommendation system still faces many challenges in heterogeneity, privacy protection security and system efficiency, and the existing research has not fully revealed the internal relationship between these problems. This paper innovatively constructs a "privacy-effectiveness-efficiency" triangular equilibrium analysis framework, uses systematic literature analysis methods, focuses on analyzing the core problems and internal relations of privacy, effectiveness and efficiency of federated recommendation systems, systematically summarizes the progress and shortcomings of current research, and proposes optimization directions, providing theoretical references for subsequent research.