LLM-Enabled Ontology Alignment to Support Complex Decision-Making in the Built Environmen

Abstract

Cross-domain knowledge integration is crucial for complex urban and architectural decision-making, with ontology engineering playing a foundational role. To address limited automation and low interpretability in ontology alignment, this study proposes an LLM-powered agent framework to support ontology reuse across built environment domains. The performance of four LLMs and three prompt strategies is explored in alignment tasks. Three specialised agents, for context generation, expert validation, and explanation synthesis, are proposed and integrated with retrieval and mapping agents. Together they form a closed-loop framework linking automatic alignment, expert review, and explanatory feedback. This agent-enhanced architecture improves the traceability and interpretability of alignment results and facilitates expert involvement and iterative refinement, offering a solid foundation for managing and reusing ontologies in complex urban-scale contexts.