Purpose/Significance To explore semantic-oriented intelligent retrieval methods，and to achieve semantic interoperability of electronic medical record(EMR)text data and enhance retrieval efficiency. Method/Process A medical terminology database and a hierarchical expansion mechanism are used to enhance retrieval. Pre-trained language models are utilized to conduct deep semantic encoding and similarity calculation on queries and medical record texts，achieving advanced semantic retrieval based on context understanding. Result/Conclusion The method can significantly improve the accuracy and recall rate，and provide a feasible solution for efficient and precise intelligent retrieval of EMR.