OntoEval-LLM: A Task-Aware Multi-Dimensional Evaluation Framework for LLM-Generated Ontologies

This work introduces OntoEval-LLM, a task-aware multi-dimensional framework for evaluating LLM-generated and extended ontologies beyond conventional exact-match metrics. Its main contribution is a seven-dimensional diagnostic profile that distinguishes semantic, structural, logical, coverage, hallucination, functional, and formatting failures, validated through fresh controlled corruptions, structural stress tests, ontology-disjoint defect diagnosis, and human-assisted review.