Ontologies are essential for structuring complex knowledge in domains such as biomedicine and healthcare, yet building reliable ontologies is difficult because class hierarchies often contain subtle subsumption errors that need careful correction. This paper addresses the issue in CASTON, a casting‑domain ontology, by proposing a lightweight yet scalable refinement workflow. The workflow first fine‑tunes an Ontology Embedding (OnT) model on CASTON to obtain vector representations of axioms. It then extracts a limited set of potentially erroneous axioms together with candidate replacements from the embedding space. These candidate pairs are fed to a Large Language Model (LLM), specifically Qwen3‑8B, which generates repair suggestions. In a small expert evaluation, the default configuration (CASTON‑fine‑tuned OnT + Qwen3‑8B) yielded high‑confidence repairs: all six high‑confidence replacements were accepted by both domain experts. The implementation is publicly available at https://github.com/HuiYang1997/OntRefine.
Review