Chinese Medical Text Entity Recognition Based On Deep Neural Network

In recent years,China's medical industry has developed rapidly in the direction of informatization,and the country has also vigorously promoted the Internetization of the medical industry.Hospitals and other medical departments across the country are also vigorously promoting hospital information systems.Electronic medical records have begun to appear in large numbers and are rapidly promoted.How to effectively use these electronic medical records and promote the medical information is the current top priority.Electronic medical records are images,text,data,images and other information recorded during a patient's medical treatment.And electronic medical records generally contain quite a lot of medical entities,which have high medical value.Therefore,many research scholars studied to identify medical entities from electronic medical records.In this article,through in-depth understanding of electronic medical records,electronic medical records usually contain patient consultation records,discharge and hospital records,doctors' medical orders and other text records.These records contain a lot of medical information such as patients' physical information,patient's condition,and patients' examination results etc.At the same time,there are many discontinuous medical entities in the medical text.These non-contiguous medical entities are ignored in entity name recognition.Based on this,this paper improves a new method CNN-GRU-LSTM based on deep neural networks.This method not only achieves better results in identifying continuous medical texts.At the same time,the effect in processing non-contiguous medical text is also higher than the existing entity naming recognition methods.In order to effectively identify non-continuous entities in medical text,based on the chain-based CRF model,we combined a convolutional neural network and GRU-CRF to build a new deep neural network.We firstly extracted the context of each word through the convolutional neural network feature learning layer,and input all the extracted feature vectors to the GRU layer.Then the context information of the sentence sequence is learned through the gated recurrent unit network,and the learned new vectors are passed to the CRF output layer.Finally,the conditional random field transfer matrices are used to predict each word label.The main contents of this article are:1.Chinese medical text is the basis for entity naming and recognition,medical data privacy and other reasons,the Chinese medical corpus is difficult to obtain,resulting in less research work on Chinese medical text.Therefore,constructing a Chinese medical corpus data set is a problem that needs to be solved.CCKS is a publicly available medical text data set,but the medical entities contained in this data set are continuous entities.We use the rules based approach to convert some continuous medical entities into non-continuous entities and obtain them through cross-validation annotation Proven Chinese non-continuous medical text data set.2.The recurrent neural network has a good recognition effect in entity recognition,and the convolutional neural network CNN have better results in processing Chinese context information,so we combine the convolutional neural network and the recurrent neural network,and constructed the CNN-GRU-CRF network model based on them.3.There are a large number of discontinuous entities in Chinese medical texts.Traditional medical named entity recognition research only focuses on continuous entities in Chinese,and it is not ideal for discontinuous entity recognition.We optimize the model by adjusting the parameters of CNN-GRU-CRF.Experiments verify that our method not only performs better than other popular entity naming recognition methods on non-continuous medical text entities,but also has a slight increase on data sets that are all continuous entities.