Entity Recognition of Chinese Electronic Medical Record Based on Gated Graph Neural Network

Named entity recognition for Chinese electronic medical records aims to identify and classify clinical terms. At present, most methods based on deep learning have weak points, such as insufficient feature extraction, ineffective identification of rare entities, and inaccurate boundary division. A novel approach to solve these problems is proposed in this paper. The approach uses graph neural networks with a multidigraph structure to capture the features that the medical dictionary offers; a convolutional neural network (CNN) is employed to obtain the boundary features between words. These two features are concatenated into one feature vector and input to the Bi-Mogrifier LSTM-Attention module. The proposed model integrates dictionary features into the deep neural network to better deal with rare clinical named entities. The Multi-Head Attention module emphasizes the correlation and dependence between current information and context information in many aspects. It weakens the weight of semantic features irrelevant to medical and clinical treatment. The experimental results show that the proposed model is superior to the mainstream sequence labeling model, and the FI value of 95.31% is achieved.