Cluster-Enhanced Federated Graph Neural Network for Recommendation

TL;DR

This paper addresses privacy in graph-based recommender systems by introducing CFedGR, a cluster-enhanced federated GNN that injects high-order collaborative signals into local user-item graphs without sharing raw data. The method clusters pretrained user representations on a central server to identify collaborative neighbors and augments local graphs while employing two communication-efficient strategies (cluster-based sampling and top-k neighbor selection). It also protects privacy via negative sampling for gradients and local differential privacy with noise addition and clipping. Experiments on FilmTrust, MovieLens-100K, and Douban show CFedGR achieves competitive accuracy with substantially reduced neighbor communication compared to prior federated GNNs that rely on extra servers. This approach offers a practical privacy-preserving path for scalable, high-order signal-aware recommendations.

Abstract

Personal interaction data can be effectively modeled as individual graphs for each user in recommender systems.Graph Neural Networks (GNNs)-based recommendation techniques have become extremely popular since they can capture high-order collaborative signals between users and items by aggregating the individual graph into a global interactive graph.However, this centralized approach inherently poses a threat to user privacy and security. Recently, federated GNN-based recommendation techniques have emerged as a promising solution to mitigate privacy concerns. Nevertheless, current implementations either limit on-device training to an unaccompanied individual graphs or necessitate reliance on an extra third-party server to touch other individual graphs, which also increases the risk of privacy leakage. To address this challenge, we propose a Cluster-enhanced Federated Graph Neural Network framework for Recommendation, named CFedGR, which introduces high-order collaborative signals to augment individual graphs in a privacy preserving manner. Specifically, the server clusters the pretrained user representations to identify high-order collaborative signals. In addition, two efficient strategies are devised to reduce communication between devices and the server. Extensive experiments on three benchmark datasets validate the effectiveness of our proposed methods.

Figures (3)

• Figure 1: Framework of the CFedGR
• Figure 2: Ablation study without collaborative neighbors
• Figure 3: Performance comparison on different values of K