Entity Recognition In Electronic Medical Records Based On Neural Network

Electronic Medical Record refers to the original information that records all the diagnostic and therapeutic processes that patients receive from entering the hospital to checking out and discharging from the hospital.In addition to a large number of medical terms and free narrative between doctors and patients,i.e.doctor's advice,taboos,ward rounds and patient's description of limb pain,it also includes test results,unit or dose,frequent abbreviated drug names,etc.Medical entity and other medical terminology acquired through the recognition of electronic medical records play vital role in accelerating the intelligent construction of computer-aided diagnosis system.Recently,more and more scholars have tried to apply deep learning model to the medical entity recognition task of electronic medical records.In this paper,the BERT model and the bidirectional long short-term memory network,which are most widely used in sequence tagging tasks,are used to carry out the research of named entity recognition in the medical field,and the experiments are carried out to verify the application of the BERT model and the bidirectional long short-term memory network.Firstly,the experimental data were tested under the BiLSTM-CRF and BERT models using the electronic medical record evaluation taks of China Conference on Knowledge Graph and Semantic Computing 2018.Named medical entities were extracted from the Chinese electronic medical record documents and their entity categories and location information were given by BIO tagging strategy.Secondly,the effect of increasing domain data sets on model recognition is discussed,and the expanded data sets are experimented with BiLSTM-CRF and BERT models.The experimental results show that the recognition effect of the model can be effectively improved by adding domain data.Finally,the experimental results on two sets of datasets are compared and analyzed with CRF baseline system and previous work.In the experiment on the evaluation data set,the BERT model has the best recognition effect,the F1 value reaches 93.93,exceeding CRF4.26%.Secondly,the BiLSTM-CRF model has a F1 value of 90.91.The experimental results of the two NER models are better than the CRF model.In the experiments on the expanded data set,the BiLSTM-CRF model showed a significant improvement,and the F1 value increased by 4.77% from 90.91 to 95.25.Compared with the previous work,the experimental results in this paper were also significantly higher than others work(the F1 values were 89.09,90.14 and 88.61,respectively).In this study,two different deep learning models were used to identify named entities in Chinese electronic medical records.The experimental results show that,compared with the traditional CRF method,BiLSTM-CRF method can realize automatic feature learning and capture the long-distance dependence of text sequence in the entity recognition of Chinese electronic medical records.At the same time,it also verifies the effectiveness of using neural network for entity recognition of electronic medical records.