Use Graph When It Needs: Efficiently and Adaptively Integrating Retrieval-Augmented Generation with Graphs

One-line Summary

EA-GraphRAG efficiently integrates retrieval-augmented generation with graph-based methods, improving accuracy and reducing latency by dynamically adapting to query complexity.

Plain-language Overview

Large language models often struggle with tasks requiring extensive knowledge due to issues like outdated information and hallucinations. A method called Retrieval-Augmented Generation (RAG) helps by pulling in external information, but it can be inefficient when dealing with fragmented data. Graph-augmented RAG (GraphRAG) was developed to improve reasoning by using structured knowledge graphs, but it can be slow and less accurate in real-world applications. The new EA-GraphRAG framework adapts dynamically to the complexity of queries, using a mix of RAG and GraphRAG techniques, resulting in better performance and speed.

Technical Details