KG-RAG: A Hybrid Retrieval Architecture for Institutional Normative Documents
Resumo
Higher education institutions deal with complex hierarchical normative documents, a scenario where traditional Retrieval-Augmented Generation (RAG) systems suffer from context fragmentation and hallucinations. Furthermore, public data management requires strictly local processing to ensure technological sovereignty. To address these limitations, this paper proposes and evaluates the Knowledge Graph RAG (KG-RAG) architecture, a hybrid pipeline applied to the documentary ecosystem of the Federal University of Western Pará (UFOPA). The model integrates an institutional ontology, Contextual Chunking, and query decomposition to isolate administrative boundaries before response generation, processing a corpus of 11,047 chunks. An empirical evaluation using the RAGAS framework demonstrated that the approach outperformed a standard hybrid baseline, achieving gains of 30.1% in Faithfulness and 35.2% in Context Recall. The results provide evidence that retrieval guided by logical relationships mitigates context loss in deep normatives, ensuring informational accuracy and privacy in closed government environments.
Referências
Edge, D., Trinh, H., Cheng, N., Bradley, J., Chao, A., Mody, A., Truitt, S., Metropolitansky, D., Ness, R. O., and Larson, J. From local to global: A graph rag approach to query-focused summarization. arXiv preprint arXiv:2404.16130 , 2024.
Es, S., James, J., Anke, L. E., and Schockaert, S. Ragas: Automated evaluation of retrieval augmented generation. In Proceedings of the 18th conference of the european chapter of the association for computational linguistics: system demonstrations. pp. 150–158, 2024.
Fensel, D., Şimşek, U., Angele, K., Huaman, E., Kärle, E., Panasiuk, O., Toma, I., Umbrich, J., and Wahler, A. Introduction: what is a knowledge graph? In Knowledge graphs: Methodology, tools and selected use cases. Springer, pp. 1–10, 2020.
Justino, A. F., Junior, A. F. L. J., and Lobato, F. M. F. A comparative study of bert models for semantic retrieval of brazilian legal precedents. In Symposium on Knowledge Discovery, Mining and Learning (KDMiLe). SBC, pp. 65–72, 2025.
Knollmeyer, S., Caymazer, O., Koval, L., Akmal, M. U., Asif, S., Mathias, S. G., and Großmann, D. Evaluating retrieval augmented generation: A comprehensive review of evaluation dimensions, question types, and application. SN Computer Science 7 (5): 576, 2026.
Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W.-t., Rocktäschel, T., et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems vol. 33, pp. 9459–9474, 2020.
Li, M., Miao, S., and Li, P. Simple is effective: The roles of graphs and large language models in knowledgegraph-based retrieval-augmented generation. In International Conference on Learning Representations. Vol. 2025. pp. 6061–6089, 2025.
Marcacini, R. M., Valverde-Rebaza, J. C., Turine, M. A., Santos, B. N., Levcovitz, S., and Rezende, S. O. Llm4gov: A privacy-preserving approach to teacher-student fine-tuning of distilled llms for the public sector. In Latin American Symposium on Digital Government (LASDiGov). SBC, pp. 273–284, 2025.
Oliveira, V. T., Lima, M. R., Teles, S., and Dias, E. S. J-kgrag: A hybrid retrieval-augmented generation architecture for legal norm understanding with knowledge graphs. In Congresso Latino-Americano de Software Livre e Tecnologias Abertas (Latinoware). SBC, pp. 202–209, 2025.
Robertson, S. and Zaragoza, H. The probabilistic relevance framework: BM25 and beyond. Vol. 4. Now Publishers Inc, 2009.
Seabra, A., Cavalcante, C., Medeiros, G., Nepomuceno, J., Sallenave, V., Ruberg, N., and Lifschitz, S. Pergunte ao bi: Uma interface conversacional para apoio à tomada de decisão usando modelos de linguagem. In Simpósio Brasileiro de Banco de Dados (SBBD). SBC, pp. 142–147, 2025.
Xavier, O. C. and da Silva Soares, A. Training-free hybrid evidence retrieval for question answering: Dynamic fusion of knowledge-graph triples and dense text embeddings. In Simpósio Brasileiro de Banco de Dados (SBBD). SBC, pp. 644–657, 2025.
Xu, Z., Cruz, M. J., Guevara, M., Wang, T., Deshpande, M., Wang, X., and Li, Z. Retrieval-augmented generation with knowledge graphs for customer service question answering. In Proceedings of the 47th international ACM SIGIR conference on research and development in information retrieval. pp. 2905–2909, 2024.
Zhu, X., Xie, Y., Liu, Y., Li, Y., and Hu, W. Knowledge graph-guided retrieval augmented generation. In 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). pp. 8912–8924, 2025.
