Retrieval-Augmented Generation (RAG) methods have attracted widespread attention since their proposal, with the core goal of enhancing large language models’ (LLMs) reasoning capabilities by incorporating external information to effectively mitigate hallucination. However, RAG approaches face challenges in complex multi-hop question answering, including incomplete information retrieval in a single pass and error propagation caused by redundant interference during iteration. To address these issues, we propose a Knowledge Graph-based Iterative Rewriting RAG method (KGIR-RAG), which constructs evidence subgraphs to filter structured knowledge from retrieved passages and utilizes a maximum spanning tree algorithm to eliminate noisy triplets. It iteratively rewrites queries by replacing multi-hop dependencies in questions with factual relations from the knowledge graph, progressively reducing reasoning complexity. Finally, a reflection-based reasoning module evaluates answer completeness to drive iteration termination. We conducted experiments on the HotpotQA and MuSiQue datasets to validate the effectiveness of each module. The results demonstrate that KGIR-RAG outperforms several recently proposed methods in both generation and retrieval performance.
LID W, YANGS, TANZ, et al. DALK: Dynamic co-augmentation of LLMs and KG to answer Alzheimer’s disease questions with scientific literature[C]//Findings of the Association for Computational Linguistics: EMNLP 2024. Stroudsburg: Association for Computational Linguistics, 2024: 2187-2205. DOI:10.18653/v1/2024.findings-emnlp.119 .
[2]
COLOMBOA. Leveraging knowledge graphs and LLMs to support and monitor legislative systems[C]//Proceedings of the 33rd ACM International Conference on Information and Knowledge Management. New York: ACM, 2024: 5443-5446. DOI:10.1145/3627673.3680268 .
[3]
XUZ T, CRUZM J, GUEVARAM, et al. Retrieval-augmented generation with knowledge graphs for customer service question answering[DB/OL].[2025-02-03].DOI: 10.1145/3626772.3661370 .
[4]
GUTIÉRREZB J, SHUY H, GUY, et al. HippoRAG: Neurobiologically inspired long-term memory for large language models[DB/OL].[2025-02-04].DOI: 10.52202/079017-1902 .
[5]
KARPUKHINV, OGUZB, MINS, et al. Dense passage retrieval for open-domain question answering[C]//Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). Stroudsburg: Association for Computational Linguistics, 2020: 6769-6781. DOI:10.18653/v1/2020.emnlp-main.550 .
[6]
ZHUX R, XIEY X, LIUY, et al. Knowledge graph-guided retrieval augmented generation[C]//Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). Stroudsburg: Association for Computational Linguistics, 2025: 8912-8924. DOI:10.18653/v1/2025.naacl-long.449 .
[7]
SHAOZ H, GONGY Y, SHENY L, et al. Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy[C]//Findings of the Association for Computational Linguistics: EMNLP 2023. Stroudsburg: Association for Computational Linguistics, 2023: 9248-9274. DOI:10.18653/v1/2023.findings-emnlp.620 .
[8]
TRIVEDIH, BALASUBRAMANIANN, KHOTT, et al. Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Stroudsburg: Association for Computational Linguistics, 2023: 10014-10037. DOI:10.18653/v1/2023.acl-long.557 .
[9]
JEONGS, BAEKJ, CHOS, et al. Adaptive-RAG: Learning to adapt retrieval-augmented large language models through question complexity[C]//Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). Stroudsburg: Association for Computational Linguistics, 2024: 7036-7050. DOI:10.18653/v1/2024.naacl-long.389 .
[10]
CHANC M, XUC P, YUANR B, et al. RQ-RAG: Learning to refine queries for retrieval augmented generation[EB/OL]. 2024: arXiv: 2404.00610.
HUANGB, WUS A, WANGW G, et al. KG-LLM-MCom: A survey on integration of knowledge graph and large language model [J]. J Wuhan Univ (Nat Sci Ed), 2024, 70(4): 397-412. DOI: 10.14188/j.1671-8836.2024.0040(Ch ).
[13]
RAM O, LEVINEY, DALMEDIGOSI, et al. In-context retrieval-augmented language models[J]. Transactions of the Association for Computational Linguistics, 2023, 11: 1316-1331. DOI:10.1162/tacl_a_00605 .
[14]
KIMG, KIMS, JEONB, et al. Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models[C]//Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: Association for Computational Linguistics, 2023: 996-1009. DOI:10.18653/v1/2023.emnlp-main.63 .
[15]
BORGEAUDS, MENSCHA, HOFFMANNJ, et al. Improving language models by retrieving from trillions of tokens[EB/OL]. 2021: arXiv: 2112.04426.
[16]
YORANO, WOLFSONT, RAM O, et al. Making retrieval-augmented language models robust to irrelevant context[EB/OL]. 2023: arXiv: 2310.01558. DOI: 10.1162/tacl_a_00605 .
[17]
SANTHANAMK, KHATTABO, SAAD-FALCONJ, et al. ColBERTv2: Effective and efficient retrieval via lightweight late interaction[C]//Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg: Association for Computational Linguistics, 2022: 3715-3734. DOI:10.18653/v1/2022.naacl-main.272 .
[18]
HUANGL, YUW J, MAW T, et al. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions[J]. ACM Transactions on Information Systems, 2025, 43(2): 1-55. DOI:10.1145/3703155 .
[19]
GAOL Y, MAX G, LINJ, et al. Precise zero-shot dense retrieval without relevance labels[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Stroudsburg: Association for Computational Linguistics, 2023: 1762-1777. DOI:10.18653/v1/2023.acl-long.99 .
[20]
GUTIÉRREZB J, SHUY H, QIW J, et al. From RAG to memory: Non-parametric continual learning for large language models[EB/OL]. 2025: arXiv: 2502.14802.
[21]
LIANGL, BOZ P, GUIZ K, et al. KAG: Boosting LLMs in professional domains via knowledge augmented generation[C]//Proceedings of the ACM on Web Conference 2025. New York:Association for Computing Machinery. 2025: 334-343. DOI:10.1145/3701716.3715240 .
[22]
EDGED, TRINHH, CHENGN, et al. From local to global: A graph RAG approach to query-focused summarization[EB/OL]. 2024: arXiv: 2404.16130.
[23]
GUOZ R, XIAL H, YUY H, et al. LightRAG: Simple and fast retrieval-augmented generation[C]//Findings of the Association for Computational Linguistics: EMNLP 2025. Stroudsburg: Association for Computational Linguistics, 2025: 10746-10761. DOI:10.18653/v1/2025.findings-emnlp.568 .
[24]
TRIVEDIH, BALASUBRAMANIANN, KHOTT, et al. MuSiQue: Multihop questions via single-hop question composition[J]. Transactions of the Association for Computational Linguistics, 2022, 10: 539-554. DOI:10.1162/tacl_a_004759 .
[25]
YANGZ L, QIP, ZHANGS Z, et al. HotpotQA: A dataset for diverse, explainable multi-hop question answering[C]//Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Brussels, Belgium. Stroudsburg: Association for Computational Linguistics, 2018: 2369-2380. DOI:10.18653/v1/d18-125 .
[26]
BILLELA, IBRAHIMB, YAHYAS. BM25 beyond query-document similarity[C]//String Processing and Information Retrieval: 26th International Symposium on String Processing and Information Retrieval (SPIRE 2019).Segovia:Springer Nature Switzerland AG, 2019:65-79. DOI: 10.1007/978-3-030-32686-9_5 .
[27]
YIBINL, LIANGD, YUC, et al. Dense retrieval with relevance-aware contrastive pre-training[C]// In Findings of the Association for Computational Linguistics. Stroudsburg: Association for Computational Linguistics, 2023:10932-10940. DOI:10.18653/v1/2023.findings-acl.695
[28]
SARTHIP, ABDULLAHS, TULIA, et al. RAPTOR: Recursive abstractive processing for tree-organized retrieval[EB/OL]. 2024: arXiv: 2401.18059.