NeurOWL Framework Enhances Reasoning in Incomplete OWL Ontologies.

OWL ontologies are crucial for formal knowledge representation and semantic reasoning, widely used in fields such as healthcare and bioinformatics. However, real-world ontologies are frequently incomplete, posing significant challenges for accurate reasoning processes. This research introduces NeurOWL, an end-to-end neuro-symbolic framework designed to tackle the problem of subsumption reasoning within incomplete OWL ontologies. It aims to determine if a candidate subsumption is semantically plausible and, if so, to provide a logically sound explanation by identifying potential missing axioms. NeurOWL achieves this by integrating both formally defined semantics and textual semantics, leveraging the power of Large Language Models and ontology embeddings. Evaluations on diverse real-world ontologies demonstrate its robust performance in jointly performing subsumption verification and abduction.