Proceedings of TUSUR University / Archive / Issue of Journal № 4, t. 28, 2025 / Graph-RAG: A Hybrid Graph-Based Retrieval for Completeness and Consistency in Regulatory and Technical Document Analysis

Graph-RAG: A Hybrid Graph-Based Retrieval for Completeness and Consistency in Regulatory and Technical Document Analysis

DOI: 10.21293/1818-0442-2025-28-4-110-120

Download article in PDF format

JATS xml

Abstract: The paper presents a study of the Graph-RAG architecture, which is an extension of Retrieval-Augmented Generation that integrates a knowledge graph for processing technical documen-tation, aimed at verifying compliance with standardization re-quirements (GOST, industry standards, technical specifica-tions), as well as verifying the fulfillment of technical specifica-tions. Validation was conducted on a corpus of 100 documents totaling 12,480 pages, including codes of practice, technical specifications, report forms, and equipment specifications. All documents were pre-annotated according to three criteria: 1) the presence of references to regulations, 2) a clear structure of re-quirements, and 3) the presence of cross-references within and between documents. The comparative analysis includes state-of-the-art architectures: Vanilla RAG, FiD, FLARE, HyDE, ModernBERT, and Graph-RAG based on Qwen3-32B. Experi-ments show that Graph-RAG outperforms its closest competitor (Qwen3 + HyDE) in accuracy (EM) by 16.8 percentage points (pp), and in factual consistency by 10.3 pp, with p < 0.001 (t-test, n = 300 queries). The paper contains 4 tables and 3 figures, and all conclusions are supported by statistically signi-ficant results.

Keywords: retrieval-augmented generation (RAG), large language models (LLMs), graph neural networks, document analysis, technical specifications, regulatory and technical documents

For citation:
Semenov R. E. Graph-RAG: A Hybrid Graph-Based Retrieval for Completeness and Consistency in Regulatory and Technical Document Analysis. Doklady Tomskogo gosudarstvennogo universiteta sistem upravleniya i radioelektroniki, 2025, vol. 28, no. 4, pp. 110–120. DOI: 10.21293/1818-0442-2025-28-4-110-120

Authors and copyright holders:

Semenov R. E. , Russian Technical University MIREA (Moscow, Russia)

1. Radyush D.V. Knowledge injection methods in question answering. Russian Technological Journal, 2025, vol.13, no. 3, pp. 21–43 (in Russ.).
2. Gao L. Precise zero-shot dense retrieval without relevance labels. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, Toronto, Canada, Curran Associates, Inc., 2023, vol. 1, pp. 1762–1777.
3. Edge D., Trinh H., Cheng N., Bradley J., Chao A., et al. From Local to Global: A GraphRAG Approach to Query Focused Summarization. arXiv preprint, 2024, DOI: 10.48550/arXiv.2404.16130.
4. Lewis P. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Advances in Neural Information Processing Systems, 2020, vol. 33, pp. 9459–9474.
5. Zharov V.V., Sarayev P.V. Sozdaniye grafovoy RAG-sistemy po tsifrovomu pravu Rossiyskoy Federatsii [Creation of a graph RAG-system for the digital law of the Russian Federation]. Vestnik nauki, 2025, vol. 1, no. 6 (87), pp. 1395–1414 (in Russ.).
6. Xu Zh., Cruz M.J., Guevara M., Wang T., Deshpande M., Wang X., Li Zh. Retrieval-augmented generation with knowledge graphs for customer service question answering. Proceedings of the 47th international ACM SIGIR conference on research and development in information retrieval, Washington DC, USA: ACM, 2024, pp. 2905–2909.
7. Cai Y., Guo Zh., Pei Y., Bian W., Zheng W. SimGRAG: Leveraging similar subgraphs for knowledge graphs driven retrieval-augmented generation. Findings of the Association for Computational Linguistics: ACL 2025, Vienna, Austria, ACL, 2025, pp. 3139–3158.
8. Peng B., Zhu Y., Liu Y., Bo X., Shi H., Hong Ch., Zhang Y., Tang S. Graph retrieval-augmented generation: A survey. ACM Transactions on Information Systems, 2024, vol. 44, no. 2, pp. 1–52.
9. Chen J., Xiao Sh., Zhang P., Luo K., Lian D., Liu Zh. M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation. Findings of the Association for Computational Linguistics: ACL 2024, Bangkok, Thailand, ACL, 2024, pp. 2318–2335.
10. Douze M., Guzhva A., Deng Ch., Johnson J., Szilvasy G., Mazaré P.-E. The faiss library. IEEE Transactions on Big Data, 2025, vol. 12, no. 2, pp. 346–361.
11. Santhanam K., Khattab O., Potts Ch., Zaharia M. PLAID: an efficient engine for late interaction retrieval. Proceedings of the 31st ACM International Conference on Information & Knowledge Management, NY, United States, ACM, 2022, pp. 1747–1756.
12. Santhanam K., Khattab O., Saad-Falcon J., Potts Ch., Zaharia M. Colbertv2: Effective and efficient retrieval via lightweight late interaction. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Seattle, United States, ACL, 2022, pp. 3715–3734.
13. Izacard G., Grave E. Leveraging passage retrieval with generative models for open domain question answering. Proceedings of the 16th conference of the European chapter of the association for computational linguistics: main volume, Online, ACL, 2021, pp. 874–880.
14. Jiang Zh., Xu F., Gao L., Sun Zh., Liu Q., Dwivedi-Yu J., Yang Y., Callan J., Neubig G. Active retrieval augmented generation. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, Singapore, ACL 2023, pp. 7969–7992.
15. Warner B., Chaffin A., Clavié B., Weller O. et al. Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, Vienna, Austria, ACL, 2025, vol. 1, pp. 2526–2547.
16. He P., Gao J., Chen W. Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding sharing. arXiv preprint arXiv:2111.09543, 2021, DOI: 10.48550/arXiv.2111.09543.
17. Bahr L., Wehner Ch., Wewerka J., Bittencourt J., Schmid U., Daub R. Knowledge graph enhanced retrieval-augmented generation for failure mode and effects analysis. Journal of Industrial Information Integration, 2025, vol. 45, pp. 100807.