LLM-Assisted Ontology Engineering and Construction of a French Legal Knowledge Graph

2026-07-27

AI Summary

This study addresses the challenge of structuring complex French regulatory texts on repairability by proposing a two-stage large language model–assisted approach. The method first performs open information extraction fused with embeddings grounded in the SEMLEG core ontology to normalize labels and induce candidate properties; it then leverages these results to guide closed-world triple extraction across the full corpus, constructing an RDF knowledge graph. Innovatively integrating open and closed extraction paradigms, this work is the first in the legal domain to systematically induce and validate predicate signatures with respect to their domain–range constraints. Experimental results demonstrate robust structured output, near-complete class alignment, and a significant reduction in entity and predicate redundancy—introducing new attributes in fewer than 20% of extracted triples—while uncovering novel semantic combinations of existing predicates.

Abstract

Maintenance regulations are complex legal texts that are difficult to exploit when addressing a specific case and challenging to integrate into operational systems. This paper presents a two-stage LLM-assisted workflow for French maintenance regulations: ontology engineering from a SEMLEG-based core ontology, followed by construction of an ontology-grounded French legal knowledge graph. The first stage consists in the open extraction of typed entities and triples from a stratified corpus sample, the normalization of labels through embedding-based fusion, and the induction of candidate object properties with their signature (domain and range). The second stage uses the resulting ontology to guide the closed extraction of triples and RDF graph construction over the full corpus. Experiments with GPT-4.1 and mistral-large-2512 show robust structured outputs, near-complete class alignment, and a substantial reduction of duplicated entities and predicates after fusion. Fewer than 20% of triples introduce unseen properties, while lower exact signature compliance reveals new domain-range combinations for existing predicates. These results point to predicate normalization and the validation of newly observed relation signatures as key refinement steps for industrial maintenance settings.

Problem

• legal knowledge graph
• ontology engineering
• maintenance regulations
• French legal texts
• structured legal data

Innovation

• LLM-assisted ontology engineering
• legal knowledge graph
• embedding-based fusion
• predicate normalization
• relation signature induction