Research on Named Entity Recognition of Medical Records Combining Confrontation and Attention Mechanism

Named entity recognition for Chinese electronic medical records is the key step of data mining and information extraction, so how to improve the accuracy of related named entity recognition model is very important. Based on the baseline model BiLSTM-CRF, a named entity recognition method for Chinese electronic medical records (EMRs) is proposed, which incorporates antagonistic training and attention mechanism. In this method, perturbation factors are added into the model embedding layer through adversarial training to generate adversarial samples for model training. In the feature extraction layer, BiLSTM module and multi-head self-attention module are combined to capture the temporal features, context information and global feature representation of text sequences more comprehensively. Experiments on the CCKS 2019 dataset show that the proposed method can effectively improve the recognition performance and robustness of the model.