Shared Embedding Spaces for AEC Ontologies: An Empirical Study of Discovery and Alignment

Abstract: The adoption of ontologies in the AEC domain is increasing, yet ontology reuse remains constrained because identifying suitable ontologies and concepts is still largely manual. This paper investigates whether shared embedding spaces can capture semantic relatedness across AEC ontologies, thereby supporting ontology discovery and alignment. We test three hypotheses regarding expert-aligned similarity, the spatial proximity of related classes and the superiority of structure-aware over purely lexical models. We embedded several widely used AEC ontologies and evaluated ontology and class-level similarities. Results suggest that embeddings operationalize expert intuition and lower barriers to ontology reuse.