Pitfalls in AI-generated ontologies: Strategies for detection and mitigation

Detecting pitfalls in ontologies aim to identify modeling patterns that can lead to inconsistencies, redundancy, or poor semantic design. Ontology design patterns have been proposed in the community and a number of tools and libraries rely on them to detect potential issues occurring during ontology modeling, but they were largely developed before the widespread use of large language models (LLMs) in ontology engineering. As a result, they do not capture a new class of artifacts that frequently appear in LLM-generated ontologies. In this paper, we introduce a complementary set of ontology pitfalls targeting patterns that are not necessary produced by human ontology engineers but commonly arise from LLM-assisted ontology generation. These include semantic redundancies, hierarchy conflicts, and property modeling artifacts. We implement detection strategies combining structural analysis, lexical similarity, and LLM-based evaluation. We evaluate the proposed approach on two ontologies generated by the Ontology Toolkit ontology generation system, showing how the new pitfalls help identify modeling issues specific to LLM-generated ontologies. Our code and data are released

at https://github.com/D2KLab/Ontology-Pitfalls-Detector.