To build a knowledge map based on the data set of TCM formulas，so as to systematically display formula entities and their relationships.Methods:Firstly，the normalized process of formula data processing and knowledge mapping were established to obtain the data set of formulas，and then the optimal model was selected from four commonly used named entity recognition models for entity extraction.Finally，the Neo4j graph database was used to build the knowledge map.Results:The bi-directional encoder representations from transformers-bi-directional long short-term memory-conditional random field(BERT-BiLSTM-CRF) model was finally selected to extract the medical entities such as symptoms，disease names of Chinese and Western medicines，and TCM syndromes from the data set，with an average F1 value of 90.55%.A normalized dataset and a knowledge map of TCM formulas were established.Conclusion:The medical entities extracted by this method provide a data basis for the clinical practice and scientific research of TCM to systematically display formula entities and their relationships.The established knowledge map realizes the knowledge retrieval of TCM formulas，which not only helps to discover the potential knowledge and internal relations in formula data but also lays a solid foundation for information integration and knowledge discovery，thus promoting the modernization of TCM.