Named Entity Recognition Of Chinese Electronic Medical Records Based On Deep Learning

Electronic medical records,a digital version of a patient’s history stored in text format,are shaping our healthcare landscape in a new way,where we can gather all the information in one place for healthcare providers to use.The automatic extraction of this knowledge from electronic medical records requires the automatic identification of named entities and their relationships that are closely related to the patient’s health in the electronic medical records.In recent years,the application of natural language processing and information extraction techniques to electronic medical records(EMRs)has attracted wide attention.A named entity is a word or phrase that clearly identifies an individual from a group of other individuals with similar properties.Examples of named entities include agency names,personnel names,location names,etc.Proteins,genes,disease names,and drugs in the field of biomedicine.Named entity recognition is the process of locating and classifying named entities in text into predefined entity categories.Using existing models to identify medical entities from electronic medical records has proved to be a challenging task because most electronic medical records are written in a hurry and are incompatible with preprocessing.In addition,incomplete syntax,numerous abbreviations,and units after numerical values make the recognition task more complex.Standard natural language processing tools are not effectively implemented when applied to electronic medical records because the entity terms of standard natural language processing are not designed for the medical field.Therefore,it is necessary to study an effective EMR entity recognition method.The research content of this paper is to use deep learning algorithm to complete the named entity recognition of Chinese electronic medical records,and to use transfer learning algorithm to solve the problem of insufficient training data.BILSTM and IDCNN,two widely used models with good performance,were selected by deep learning algorithm.Based on BERT,BERT was combined with BILSTM(or IDCNN)+ CRF model.BERT model parameters of pre-training were loaded,and fine-tuning was performed on experimental data to complete the identification task of Chinese electronic medical record named entity with small training data.In the experiment,we compared the performance of BILSTM-CRF and IDCNN-CRF models,and compared the experimental results before and after BERT and ALBERT using the pretraining model.Precision,recall and F-score were selected as evaluation metrics.The experimental results showed that BILSTM performed better than IDCNN.The pre-training model significantly improved the experimental results,and ALBERT performed better than BERT.The best model combination was ALBERT-BILSTM-CRF.