Named entity recognition of Chinese electronic medical records based on BERT-BiLSTM-CRF

Aiming at the low accuracy of Chinese electronic medical record named entity recognition, this paper attempts to apply the Bert bilstm CRF model to Chinese electronic medical record named entity recognition. First, the BERTmodel is used to encode sentences of any length into vectors of equal length, and then the encoded word vectors are input into the BiLSTM model to obtain the semantic features of the context. Finally, the CRF layer is used to annotate the score matrix input by the BiLSTM model in sequence, and the result with the largest probability is output. The F1 value of the model applied to the 2020 CCKS competition data set and the independently annotated family history data set reached 78% and 93%, respectively. Compared with the CCMNN algorithm used in the 2020 National Knowledge Mapping and Semantic Computing Conference competition, the F1 value of entity recognition of the CCKS competition data set increased by 17%. The results show that the BERT-BiLSTM-CRF model has strong generalization ability for Chinese electronic medical record named entity recognition, and improves the accuracy of Chinese electronic medical record named entity recognition.