FedGNN: Federated Graph Neural Network for Privacy-Preserving Recommendation

Abstract: Graph neural network (GNN) is widely used for recommendation to model high-order interactions between users and items. Existing GNN-based recommendation methods rely on centralized storage of user-item graphs and centralized model learning. However, user data is privacy-sensitive, and the centralized storage of user-item graphs may arouse privacy concerns and risk. In this paper, we propose a federated framework for privacy-preserving GNN-based recommendation, which can collectively train GNN models from decentralized user data and meanwhile exploit high-order user-item interaction information with privacy well protected. In our method, we locally train GNN model in each user client based on the user-item graph inferred from the local user-item interaction data. Each client uploads the local gradients of GNN to a server for aggregation, which are further sent to user clients for updating local GNN models. Since local gradients may contain private information, we apply local differential privacy techniques to the local gradients to protect user privacy. In addition, in order to protect the items that users have interactions with, we propose to incorporate randomly sampled items as pseudo interacted items for anonymity. To incorporate high-order user-item interactions, we propose a user-item graph expansion method that can find neighboring users with co-interacted items and exchange their embeddings for expanding the local user-item graphs in a privacy-preserving way. Extensive experiments on six benchmark datasets validate that our approach can achieve competitive results with existing centralized GNN-based recommendation methods and meanwhile effectively protect user privacy.

• The paper introduces a decentralized learning framework that trains GNN models using local user data, eliminating the need for centralized storage.
• It employs advanced privacy-preserving methods including local differential privacy and pseudo interacted items to secure sensitive user information.
• Experimental results on six benchmark datasets show that FedGNN achieves competitive performance while robustly protecting user privacy.

FedGNN: Federated Graph Neural Network for Privacy-Preserving Recommendation

The paper "FedGNN: Federated Graph Neural Network for Privacy-Preserving Recommendation" addresses challenges in the implementation of graph neural networks (GNNs) for recommendation systems, particularly focusing on privacy concerns associated with centralized data storage. The proposed solution, FedGNN, is a federated learning framework designed to allow decentralized training of GNN models without compromising user data privacy.

Key Contributions

1. Federated Learning Framework: The study introduces a federated learning approach where GNN models are trained across decentralized user data. This method leverages local data on user devices to build and update models, thereby ensuring that sensitive user-item interaction data does not need to reside on a centralized server.
2. Privacy-Preserving Techniques: To further protect user data during the aggregation of local model gradients, FedGNN employs several advanced privacy-preserving techniques:
   • Local Differential Privacy (LDP): By applying LDP to local model gradients before they are shared with the centralized server, the framework significantly reduces the risk of data leakage.
   • Pseudo Interacted Items: The approach samples randomly chosen items that users have not interacted with and generates gradients for these items to obscure the real user behavior from the server.
3. High-Order Interaction Modeling: FedGNN effectively models high-order interactions between users and items by proposing a privacy-preserving graph expansion method. This method securely identifies neighboring users with shared item interactions to enrich the local user-item graphs, thus enhancing the GNN model's ability to learn complex user-item interactions without compromising privacy.
4. Experimental Validation: Extensive experiments conducted on six benchmark datasets (MovieLens-100K, MovieLens-1M, MovieLens-10M, Flixster, Douban, and YahooMusic) demonstrated that FedGNN achieves competitive results when compared to traditional, centralized GNN-based recommendation methods. Importantly, it does so while ensuring effective privacy protection.

Implications and Future Directions

The practical implications of FedGNN are significant in environments where regulatory requirements or user expectations necessitate stringent data privacy. By maintaining user data on local devices and employing privacy-preserving measures, recommendation systems can be both effective and compliant with privacy standards such as GDPR.

Theoretically, this approach advances the field of federated learning by providing a viable framework for incorporating complex model types like GNNs, which are traditionally challenging to scale across decentralized data. The ability to utilize high-order interaction data in a privacy-preserved manner could inspire further research into similar architectures for other machine learning applications.

Looking forward, future work could explore optimizations to further reduce communication overhead between user devices and servers, enabling more efficient federated learning for resource-constrained environments. Additionally, broader investigations into varied graph structures and user-item interaction types could expand the applicability of FedGNN in diverse fields beyond recommendations, such as in social networks, financial services, and personalized healthcare systems.

Overall, FedGNN presents a novel and important stride towards integrating privacy considerations into the development of powerful, graph-based recommendation algorithms.