Graph-R1: Towards Agentic GraphRAG Framework via End-to-end Reinforcement Learning
This work addresses challenges in retrieval-augmented generation for AI systems, offering an incremental improvement over GraphRAG methods.
The paper tackles the limitations of GraphRAG methods, such as high construction costs and fixed retrieval, by proposing Graph-R1, an agentic framework using end-to-end reinforcement learning, which outperforms existing methods in reasoning accuracy, retrieval efficiency, and generation quality on standard datasets.
Retrieval-Augmented Generation (RAG) mitigates hallucination in LLMs by incorporating external knowledge, but relies on chunk-based retrieval that lacks structural semantics. GraphRAG methods improve RAG by modeling knowledge as entity-relation graphs, but still face challenges in high construction cost, fixed one-time retrieval, and reliance on long-context reasoning and prompt design. To address these challenges, we propose Graph-R1, an agentic GraphRAG framework via end-to-end reinforcement learning (RL). It introduces lightweight knowledge hypergraph construction, models retrieval as a multi-turn agent-environment interaction, and optimizes the agent process via an end-to-end reward mechanism. Experiments on standard RAG datasets show that Graph-R1 outperforms traditional GraphRAG and RL-enhanced RAG methods in reasoning accuracy, retrieval efficiency, and generation quality.