Towards Open-World Retrieval-Augmented Generation on Knowledge Graph: A Multi-Agent Collaboration Framework

TL;DR

This work tackles open-world KG-based QA by removing the need for predefined anchor entities. It introduces AnchorRAG, a multi-agent collaboration framework with a Predictor for anchor grounding, parallel Retriever agents for multi-hop exploration, and a Supervisor for strategy synthesis and final answer generation. The method combines FAISS-based anchor candidate retrieval with SBERT for disambiguation, LLM-driven relation/entity ranking, and an evidence-validation step that stores validated subgraphs in shared memory, with a lightweight hybrid variant for efficiency. Empirical results across four public KGQA benchmarks show state-of-the-art performance, particularly on complex multi-hop reasoning tasks, and demonstrate robustness to open-world noise, at the cost of higher compute, which is mitigated by the efficient variant. Overall, AnchorRAG advances open-world reasoning on knowledge graphs by decoupling anchor grounding from retrieval and enabling parallel, evidence-grounded exploration.

Abstract

Large Language Models (LLMs) have demonstrated strong capabilities in web search and reasoning. However, their dependence on static training corpora makes them prone to factual errors and knowledge gaps. Retrieval-Augmented Generation (RAG) addresses this limitation by incorporating external knowledge sources, especially structured Knowledge Graphs (KGs), which provide explicit semantics and efficient retrieval. Existing KG-based RAG approaches, however, generally assume that anchor entities are accessible to initiate graph traversal, which limits their robustness in open-world settings where accurate linking between the user query and the KG entity is unreliable. To overcome this limitation, we propose AnchorRAG, a novel multi-agent collaboration framework for open-world RAG without the predefined anchor entities. Specifically, a predictor agent dynamically identifies candidate anchor entities by aligning user query terms with KG nodes and initializes independent retriever agents to conduct parallel multi-hop explorations from each candidate. Then a supervisor agent formulates the iterative retrieval strategy for these retriever agents and synthesizes the resulting knowledge paths to generate the final answer. This multi-agent collaboration framework improves retrieval robustness and mitigates the impact of ambiguous or erroneous anchors. Extensive experiments on four public benchmarks demonstrate that AnchorRAG significantly outperforms existing baselines and establishes new state-of-the-art results on the real-world reasoning tasks.

Figures (6)

• Figure 1: Illustration of the difference between the closed-world and the open-world setting in RAG.
• Figure 2: An overview of the AnchorRAG framework, including the collaborative workflows of multi-agent exploration for question answering.
• Figure 3: Efficiency and Performance Comparison of Different Methods on the WebQSP and GrailQA Datasets, using GPT-4o-mini as the backbone LLM.
• Figure 4: The effect of the anchor identifying and knowledge retrieval on WebQSP.
• Figure 5: Performances of AnchorRAG with different search depths and widths.