Federated graph retrieval improves knowledge use across organizations
Federated Graph Retrieval-Augmented Generation
Machine Learning
Summary
Many organizations keep valuable knowledge in separate graph databases but cannot share them due to privacy rules. To solve this, the paper develops a way for connected groups to find and combine relevant information without exposing their whole data. They create a system that uses small, safe summaries to locate useful knowledge across the network and then rebuild detailed evidence for generating answers. Their approach performs better than existing methods in both finding and using information from different sources.
What this means in practice
- •For enterprise data teams: Retrieve and combine sensitive knowledge graphs across company divisions without sharing full local data, improving internal knowledge access.
- •For cloud service providers: Offer federated graph retrieval features that enhance language model answers over distributed customer data while maintaining privacy.$Commercial implications: Enables privacy-preserving, distributed knowledge services for businesses, supporting new secure AI products in multi-tenant cloud environments.
Authors
Xiaobao Wang, Renda Han, Zhiliang Xia, Meng Liu, Renxiang Guan, Di Jin
Abstract
Graph Retrieval-Augmented Generation (GraphRAG) leverages centralized knowledge graphs to enhance large language models with structured knowledge retrieval. However, real-world knowledge is inherently distributed across multiple organizations rather than being centrally accessible. Privacy constraints further prohibit sharing entire local knowledge graphs, preventing GraphRAG from effectively retrieving and utilizing distributed knowledge. To this end, we propose Federated GraphRAG (FedGRAG), a unified framework for distributed graph retrieval and knowledge-enhanced generation over multiple clients with knowledge graphs. Specifically, we address two fundamental challenges in federated GraphRAG. First, the absence of a global knowledge graph makes unified retrieval impossible. To address this, we develop a Federated Knowledge Routing mechanism that efficiently identifies relevant clients through lightweight knowledge indices without exposing local graph structures. Second, local graph memory cannot be shared directly because it contains sensitive entities and knowledge information. We therefore propose a controlled-disclosure evidence construction method. It extracts retrieved evidence subgraphs and converts them into an evidence abstract, enabling evidence conversion and reconstruction on the server. Extensive experiments demonstrate that Federated GraphRAG consistently outperforms existing baselines in both retrieval and downstream generation tasks, establishing a new paradigm for deploying GraphRAG in federated environments.