Purpose/Significance To address the issues of complex structure，low query efficiency and high requirements for semantic understanding of medical knowledge graphs，an intelligent graph pruning method is proposed. Method/Process A semantic pruning strat-egy based on knowledge base question answering and large language model is adopted，a structural pruning method based on parametersand ontologies is designed. A subgraph discovery workflow is established for natural language question answering and graph structure ex-ploration. Result/Conclusion The method has good accuracy，scalability and cross-lingual processing capabilities. It has been applied to MedKaaS platform，facilitating precise querying and knowledge services of massive medical science and technology resources.