OntoLearner Library Unifies Ontology Learning with LLMs and Benchmarking

Ontology learning (OL), the process of automatically constructing structured knowledge models from text, has long been hampered by fragmented research, inconsistent methods, and a lack of shared infrastructure for evaluation. This has made systematic progress difficult. To address these challenges, researchers have developed OntoLearner, a modular Python library. This first-of-its-kind framework unifies access to ontologies, provides pipelines for LLM-driven learning, and offers standardized benchmarking capabilities. It includes 180 machine-readable ontologies across 22 domains and ready-to-use datasets for key OL tasks like term typing, taxonomy discovery, and non-taxonomic relation extraction. A large-scale empirical study conducted using OntoLearner, evaluating 22 retrieval models and 12 LLMs, revealed a significant insight: the primary bottleneck in ontology learning is not merely model capability or size, but rather the structural mismatch between how models encode knowledge and how ontologies organize it. Failure modes tend to scale with the inherent complexity of the ontology itself. This finding redefines the central challenge of OL and underscores the value of OntoLearner's comprehensive benchmarking.