Graph RAG for Legal Norms: A Hierarchical, Temporal and Deterministic Approach (2505.00039v3)

Abstract: This article proposes an adaptation of Graph Retrieval-Augmented Generation (Graph RAG) specifically designed for the analysis and comprehension of legal norms. Legal texts are characterized by a predefined hierarchical structure, an extensive network of references and a continuous evolution through multiple temporal versions. This temporal dynamism poses a significant challenge for standard AI systems, demanding a deterministic representation of the law at any given point in time. To address this, our approach grounds the knowledge graph construction in a formal, FRBRoo-inspired model that distinguishes abstract legal works from their concrete textual expressions. We introduce a multi-layered representation of Temporal Versions (capturing date-specific changes) and Language Versions (capturing linguistic variations). By modeling normative evolution as a precise sequence of these versioned entities, we enable the construction of a knowledge graph that serves as a verifiable "ground truth". This allows LLMs to generate responses based on accurate, context-aware, and point-in-time correct legal information, overcoming the risk of temporal inaccuracies. Through a detailed analysis of this formal Graph RAG approach and its application to legal norm datasets, this article aims to advance the field of Artificial Intelligence applied to Law, creating opportunities for more effective and reliable systems in legal research, legislative analysis, and decision support.