SBV-LawGraph: A Hybrid RAG Approach Integrating Knowledge Graph for the State Bank of Vietnam Legal Documents

Abstract

While Retrieval-Augmented Generation (RAG) pipelines often demonstrate strong performance in general settings, they struggle with legal texts, where interpreting structure and relationships between laws is crucial. To address this, we introduce SBV-LawGraph – a dual-retrieval framework designed specifically for Vietnamese legal documents. It combines semantic retrieval with graph-based reasoning by integrating two modules: a Legal Retrieval module using sparse–dense reranking for textual accuracy, and a Relationship Retrieval module that traverses a curated Legal Knowledge Graph to capture links such as amendments, citations, and definitions. This design enables SBV-LawGraph to generate responses that are not only relevant but also structurally grounded, addressing limitations of standard RAG systems. Evaluations on the ALQAC2025 and SBV Legal Questions datasets show it consistently outperforms strong baselines, highlighting its effectiveness for precise and explainable legal QA. Keywords: Retrieval-Augmented Generation, Knowledge Graph, Natural Language Processing, Question-Answering Systems

Cite this paper

@inproceedings{le2026sbvlawgraph,
  author    = {Phan, Khoa and Le, Xuan-Bach and Quan, Tho},
  title     = {SBV-LawGraph: A Hybrid RAG Approach Integrating Knowledge Graph for the State Bank of Vietnam Legal Documents},
  booktitle = {Proceedings of the 18th Asian Conference on Intelligent Information and Database Systems (ACIIDS 2026)},
  series    = {Lecture Notes in Computer Science},
  pages     = {235--250},
  publisher = {Springer},
  isbn      = {978-981-92-0071-9},
  year      = {2026},
  address   = {Kaohsiung, Taiwan},
  doi       = {10.1007/978-981-92-0071-9_16}
}