Elamur Research⌕
☰
home / papers / 2026 / TopoRAG: Graph-based RAG via Topology-aware Approximate Nearest Neighbor Search

TopoRAG: Graph-based RAG via Topology-aware Approximate Nearest Neighbor Search

Paper · 2026 · Tianhao Wu and Siqiang Luo

Summary

Graph-based RAG systems like GraphRAG and LightRAG usually find evidence in two steps. A plain vector ANN lookup first picks the top-k entities. Their graph neighbours are then added by heuristic expansion. Because the graph is ignored in the first step, the expansion tends to pull in weakly related entities and high-degree hubs, which inflates prompts and lowers precision.

Wu and Luo (NTU Singapore) recast retrieval as a diameter-constrained ANN problem. The goal is to choose k entities close to the query embedding whose induced subgraph has diameter at most D. This lets semantic similarity and graph structure be handled in one search. TopoRAG runs on any pre-built entity graph and adds three practical modules:

  • extra edges from entity co-occurrence inside chunks;
  • splitting complex queries into sub-queries;
  • hybrid entity-plus-chunk retrieval.

On several QA benchmarks, including MultiHopRAG, it outperforms vanilla RAG and graph-based baselines such as ArchRAG in accuracy while reducing redundant context. It can replace neighbour expansion in existing graph RAG pipelines without changing how the graph is built.

Key points

  • Formulation: For a graph G = (V, E) with node embeddings and a query embedding h_q, TopoRAG selects a fixed-size node set S_q (size k) that is near the query and whose induced subgraph diameter is bounded by D. D is kept small and matched to typical reasoning hop length (about 2–3 hops in knowledge graphs). Raising D improves accuracy up to a point, after which subgraphs become loose and noisy.
  • Ablations (Tables 4–5): Diameter filtering applied inside the search ("in-filter", with pruning) is more effective than post-filtering. Removing query decomposition, graph augmentation, or chunk-level retrieval each lowers performance, and chunk retrieval notably helps multi-hop questions.
  • Comparison with ArchRAG on MultiHopRAG: TopoRAG is more accurate at every tested number of ANN candidates k. Larger k raises accuracy for both systems but also increases time per query and token usage. Exact headline figures fall in the part of the extracted text that was omitted.
  • Limitations: Compact subgraphs may miss long-range or global dependencies. The method also needs a static, pre-constructed entity graph and does not support dynamically built graphs.
  • Publication: Findings of ACL 2026, pp. 34097–34108, funded by the National Research Foundation Singapore (NRF-F-CRP-2024-0005).

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Nanyang Technological UniversitySingapore · Singapore184 works · 27,484 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

Cite

Cite as Tianhao Wu and Siqiang Luo. 2026. TopoRAG: Graph-based RAG via Topology-aware Approximate Nearest Neighbor Search. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1703

@inproceedings{wu2026toporag,
  title = {{TopoRAG: Graph-based RAG via Topology-aware Approximate Nearest Neighbor Search}},
  author = {Tianhao Wu and Siqiang Luo},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1703},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1703}
}

References

28 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Kai Guo, Xinnan Dai, Shenglai Zeng, Harry Shomer, Haoyu Han, Yu Wang, and Jiliang Tang. 2025. Beyond static retrieval: Opportunities and pitfalls of iterative retrieval in GraphRAG. arXiv preprint arXiv:2509.25530. — [arXiv:2509.25530](https://arxiv.org/abs/2509.25530) find
  2. Balabhaskar Balasundaram and F Mahdavi Pajouh. 2013. Graph theoretic clique relaxations and applications. Handbook of Combinatorial Optimization, pages 1559–1598. find
  3. Zirui Guo, Lianghao Xia, Yanhua Yu, Tu Ao, and Chao Huang. 2024. LightRAG: Simple and fast retrieval-augmented generation. arXiv preprint arXiv:2410.05779.
  4. Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang. 2020. Retrieval augmented language model pre-training. In International Conference on Machine Learning, pages 3929–3938. PMLR.
  5. Xiaoxin He, Yijun Tian, Yifei Sun, Nitesh Chawla, Thomas Laurent, Yann LeCun, Xavier Bresson, and Bryan Hooi. 2024. G-Retriever: Retrieval-augmented generation for textual graph understanding and question answering. Advances in Neural Information Processing Systems, 37:132876–132907.
  6. Piotr Indyk and Rajeev Motwani. 1998. Approximate nearest neighbors: Towards removing the curse of dimensionality. In Proceedings of the thirtieth annual ACM Symposium on Theory of Computing, pages 604–613. find
  7. Herve Jegou, Matthijs Douze, and Cordelia Schmid. 2010. Product quantization for nearest neighbor search. IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(1):117–128. find
  8. Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Ye Jin Bang, Andrea Madotto, and Pascale Fung. 2023. Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12):1–38.
  9. Bernal Jimenez Gutierrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su. 2024. HippoRAG: Neurobiologically inspired long-term memory for large language models. Advances in Neural Information Processing Systems, 37:59532–59569.
  10. Soohyeong Kim, Seok Jun Hwang, JungHyoun Kim, Jeonghyeon Park, and Yong Suk Choi. 2025. Re-GraphRAG: Reorganizing fragmented knowledge graphs for multi-perspective retrieval-augmented generation. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 5426–5443. find
  11. Tomáš Kočiskỳ, Jonathan Schwarz, Phil Blunsom, Chris Dyer, Karl Moritz Hermann, Gábor Melis, and Edward Grefenstette. 2018. The NarrativeQA reading comprehension challenge. Transactions of the Association for Computational Linguistics, 6:317–328.
  12. Yi-Xiu Kong, Gui-Yuan Shi, Rui-Jie Wu, and Yi-Cheng Zhang. 2019. k-core: Theories and applications. Physics Reports, 832:1–32. find
  13. Jakub Lála, Odhran O’Donoghue, Aleksandar Shtedritski, Sam Cox, Samuel G Rodriques, and Andrew D White. 2023. PaperQA: Retrieval-augmented generative agent for scientific research. arXiv preprint arXiv:2312.07559.
  14. Tommaso Lanciano, Atsushi Miyauchi, Adriano Faz-zone, and Francesco Bonchi. 2024. A survey on the densest subgraph problem and its variants. ACM Computing Surveys, 56(8):1–40. find
  15. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems, 33:9459–…
  16. Hui Li, Shiyuan Deng, Xiao Yan, Xiangyu Zhi, and James Cheng. 2025. SAQ: Pushing the limits of vector quantization through code adjustment and dimension segmentation. Proceedings of the ACM on Management of Data, 3(6):1–25. find
  17. Wen Li, Ying Zhang, Yifang Sun, Wei Wang, Mingjie Li, Wenjie Zhang, and Xuemin Lin. 2019. Approximate nearest neighbor search on high dimensional data—experiments, analyses, and improvement. IEEE Transactions on Knowledge and Data Engineering, 32(8):1475–1488. find
  18. Ting Liu, Andrew Moore, Ke Yang, and Alexander Gray. 2004. An investigation of practical approximate nearest neighbor algorithms. Advances in Neural Information Processing Systems, 17. find
  19. Linhao Luo, Zicheng Zhao, Gholamreza Haffari, Dinh Phung, Chen Gong, and Shirui Pan. 2025. GFM-RAG: Graph foundation model for retrieval augmented generation. In Proceedings of the Thirty-ninth Annual Conference on Neural Information Processing Systems, pages 1–35. find
  20. Yu A Malkov and Dmitry A Yashunin. 2018. Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(4):824–836.
  21. Julieta Martinez, Shobhit Zakhmi, Holger H Hoos, and James J Little. 2018. LSQ++: Lower running time and higher recall in multi-codebook quantization. In Proceedings of the European Conference on Computer Vision (ECCV), pages 491–506. find
  22. Alistair Moffat and Justin Zobel. 1996. Self-indexing inverted files for fast text retrieval. ACM Transactions on Information Systems (TOIS), 14(4):349–379. find
  23. Flavia Moser, Recep Colak, Arash Rafiey, and Martin Ester. 2009. Mining cohesive patterns from graphs with feature vectors. In Proceedings of the 2009 SIAM International Conference on Data Mining, pages 593–604. SIAM. find
  24. Naoki Ono and Yusuke Matsui. 2023. Relative NN-Descent: A fast index construction for graph-based approximate nearest neighbor search. In Proceedings of the 31st ACM International Conference on Multimedia, pages 1659–1667. find
  25. Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh Khanna, Anna Goldie, and Christopher D Manning. 2024. RAPTOR: Recursive abstractive processing for tree-organized retrieval. In The Twelfth International Conference on Learning Representations.
  26. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement learning. Advances in Neural Information Processing Systems, 36:8634–8652.
  27. Zulun Zhu, Tiancheng Huang, Kai Wang, Junda Ye, Xinghe Chen, and Siqiang Luo. 2025. Graph-based approaches and functionalities in retrieval-augmented generation: A comprehensive survey. arXiv preprint arXiv:2504.10499. — [arXiv:2504.10499](https://arxiv.org/abs/2504.10499) find
  28. Yixuan Tang and Yi Yang. 2024. MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries. In First Conference on Language Modeling.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wu-2026-toporag-graph-based-rag-topology/