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

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: G{é}nesis Montenegro, Mokhtar Boumedyen Billami, Catherine Faron, Fabien Gandon, Pierre Monnin
• Venue: arXiv: cs.AI
• Published: 2026-07-27
• Collected: 2026-07-28
• URL: http://arxiv.org/abs/2607.24551v1
• PDF: https://arxiv.org/pdf/2607.24551v1
• DOI: UNKNOWN

Plain-Language Brief

The paper describes a two-stage workflow for creating a French legal knowledge graph, specifically for maintenance regulations. The first stage involves open extraction of entities and triples from a corpus using LLM (Large Language Model), normalization of labels, and induction of candidate object properties. The second stage uses the resulting ontology to guide the closed extraction of triples and RDF graph construction over the full corpus. The experiments with GPT-4.1 and mistral-large-2512 showed promising results, including robust structured outputs, near-complete class alignment, and a reduction of duplicated entities and predicates.

What the Document Says

• The workflow presented is two-stage for French maintenance regulations.
• First stage: open extraction of typed entities and triples from stratified corpus sample, label normalization, induction of candidate object properties.
• Second stage: uses ontology for closed extraction of triples and RDF graph construction over full corpus.
• Experiments with GPT-4.1 and mistral-large-2512 show robust structured outputs, near-complete class alignment, reduction of duplicated entities and predicates.
• Fewer than 20% of triples introduce unseen properties, lower exact signature compliance reveals new domain-range combinations for existing predicates.

Key Concepts Mentioned

• Maintenance regulations
• LLM (Large Language Model)
• Ontology engineering
• SEMLEG-based core ontology
• Typed entities and triples
• RDF graph construction
• Class alignment
• GPT-4.1
• mistral-large-2512

Claims Made by the Document

• The workflow presents a robust solution for handling complex legal texts.
• GPT-4.1 and mistral-large-2512 produce structured outputs with near-complete class alignment.

Evidence Used

• Results from experiments with GPT-4.1 and mistral-large-2512.

Methods / Approach

• Open and closed extraction of entities and triples.
• Label normalization through embedding-based fusion.
• Induction of candidate object properties.

Limitations Stated by the Authors

• Predicate normalization and validation of newly observed relation signatures identified as key refinement steps for industrial maintenance settings.

Limitations Noted by the Agent

Agent interpretation — not from the source.

• The study focuses on French maintenance regulations, which may limit its applicability to other legal domains or regions.

Open Questions

• What are the specific challenges faced in industrial maintenance settings that necessitate these refinement steps?

Notes for Future Q&A

May include interpretation. Labeled where it does.

• The paper does not specify what 'exact signature compliance' means in this context.
• It is unclear whether the experiments were conducted on a representative sample of French maintenance regulations.

Raw Abstract / Excerpt

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.