Multi-Agent LLM Approach for Automated Ontology Generation
A recent investigation published on arXiv (2604.23090) delves into the automation of ontology creation from unstructured text through the use of large language models. The researchers set a baseline with a single-agent LLM, uncovering issues such as inadequate compliance with Ontology Design Patterns, excessive structural redundancy, and subpar iterative repair processes. Subsequently, they proposed a multi-agent framework that divides ontology development into four distinct roles: Domain Expert, Manager, Coder, and Quality Assurer. The evaluation of performance focused on architectural quality, assessed by diverse LLM judges, and functional usability, determined through competency question-driven SPARQL evaluations. The study utilized domain-specific insurance contracts as its data source.
Key facts
- arXiv paper 2604.23090
- Focus on automated ontology generation from unstructured text
- Single-agent LLM baseline established
- Failure modes include poor Ontology Design Pattern compliance, structural redundancy, ineffective iterative repair
- Multi-agent architecture with four roles: Domain Expert, Manager, Coder, Quality Assurer
- Evaluation via LLM judges and SPARQL query testing
- Domain-specific test data: insurance contracts
Entities
Institutions
- arXiv