Buildings are the places where occupants spend most of their time, and they hold an irreplaceable role in keeping occupants’ health. Improving the resilience of building health (BHR) can resist threats or disturbances and keep the health of both occupants and the building in a stable condition. Constructing a structured knowledge graph according to past experience is an effective way to understand the whole process of BHR including influencing factors, response actions, etc., and guide BHR improvement. However, due to the BHR being a new concept, the ontology of BHR is vague and there is limited structured data for constructing knowledge. To solve the limitations, this study proposes a framework using large language model (LLM) to empower the knowledge construction process. The first step is constructing the ontology of BHR based on the integration of bottom-up and up-down methods. CHATGPT-4o is employed to extract triples automatically because of its superior performance in natural language processing tasks. Neo4j is applied to realize knowledge graph visualization based on the extracted triples. The knowledge graph of BHR shows the potential relations between threats, the health state of the building and occupant, reasons for health state fluctuation, and response actions for BHR management, which can provide insight into understanding BHR and also offer guidance for BHR improvement.