LLM-Assisted Ontology Engineering for French Legal Knowledge Graph
A recent paper on arXiv (2607.24551) details a two-step process that uses large language models (LLM) for handling French maintenance regulations. The first step involves extracting typed entities and triples from a carefully selected sample of the text, followed by normalizing labels through an embedding-based approach and identifying potential object properties. In the second step, the created ontology enables the extraction of triples and the formation of an RDF graph for the entire dataset. When tested with GPT-4.1 and mistral-large-2512, the results showed impressive structured outputs, excellent class alignment, and a significant reduction in duplicate entities. This study addresses the complexities of integrating detailed legal documents into operational frameworks.
Key facts
- Paper arXiv:2607.24551 presents LLM-assisted workflow for French legal knowledge graph
- Two-stage process: ontology engineering then knowledge graph construction
- First stage: open extraction, label normalization, property induction
- Second stage: closed extraction and RDF graph construction
- Experiments with GPT-4.1 and mistral-large-2512
- Results show robust structured outputs and class alignment
- Reduction of duplicated entities achieved
- Focus on French maintenance regulations
Entities
Institutions
- arXiv