Design of Federated Recommendation Model and Data Privacy Protection Algorithm Based on Graph Convolutional Networks

To enhance the performance and privacy of recommender systems in distributed settings, this paper proposes a federated recommendation model based on graph convolutional networks (GCNs). Leveraging local user-item bipartite graphs, the model extracts high-order interaction features and performs parameter aggregation across clients. A Gaussian mechanism is introduced to enforce ε-differential privacy, combined with a secure aggregation protocol based on secret sharing to mitigate embedding leakage and reconstruction risks. Experimental analysis shows that the proposed multi-layer framework achieves high accuracy and stable convergence on heterogeneous datasets. The integration of privacy budget scheduling and gradient trimming further improves model robustness under attack scenarios. The results demonstrate that the model offers strong structural modeling and privacy protection capabilities, supporting personalized recommendation in high-risk environments.