Research On Chinese Electronic Medical Record Named Entity Recognition Method Based On Improved Deep Belief Networks

With the rapid development of computer-related science and technology,information technology in the medical and health field has also been widely used and rapidly improved.Under the strong support of a series of national policies,the Hospital Information System(HIS)has rapidly become popular.This has led to a huge amount of Electronic Medical Records(EMR)data recorded in narrative form.EMR also contains a large amount of medical information to be tapped.Then,Named Entity Recognition(NER)in the Natural Language Processing(NLP)field was introduced.Named entity recognition method is one of the important research directions of various researchers native and abroad.Although NER has developed for a long time,its recognition accuracy and F1-score are not able to satisfy the current needs.The main reason are that the traditional NER method of machine learning is often basically the maximization of the calculation conditional probability,and it is difficult to obtain the relevant semantics and semantic deep meaning information of similar words such as synonymous,anti and near meaning in text data.In view of this,this paper conducts NER research on EMR based on deep learning.Its main research contents are as follows:First of all,this paper conducts targeted research on various commonly used traditional named entity recognition methods native and abroad,including the underlying principles,advantages,disadvantages,and applications in reality.This paper analyzes the bottleneck of current electronic medical records named entity recognition,and proposes improved ideas for specific issues.Second,the lack of uniform and standardized electronic medical record data caused by insufficient domestic electronic medical record research.This article studies an incremental approach to fusion Chinese-language medical records.The method realizes the fusion of unstructured text medical record data without affecting the stability of the original system.The operation is simple and the efficiency is high,and no secondary correction is needed.In accordance with this method,a corpus of EMRs was formed for NER or further research.Subsequently,this paper conducts a detailed study of the Deep Belief Network(DBN),the basic model of deep learning.And for the NER research of electronic medical records,the improved DBN model is proposed to increase the part-of-speech(POS)nodes to promote the recognition effect.Based on the improved DBN,the use of word vectors as input methods not only solves the problem that the deep learning model cannot use text data as a direct input,but also can obtain semantic and sentence meaning information.At last,this paper study the improved DBN named entity recognition method.Then compared it with traditional DBN method,Maximum Entropy Markov Model(MEMM)method and Conditional Random Field(CRF)method.Record its performance on training data and test data.The experimental results show that the NER method of the improved DBN model proposed in this paper has the best effect and the F1-value reaches 91.749%.The method exceeds the CRF model by about 0.4%;it exceeds the 0.8%by the MEMM model method.The result shows that the word vector DBN model can be used to identify the Chinese electronic medical record named entity,and the improved DBN with the part-of-speech node is more effective.Afterwards,the results of the experiment were analyzed.It was found that the word semantics could be obtained by using the word vector.The input of the DBN model can be used to identify the Chinese electronic medical records.And the use of the improved DBN model with POS has better recognition effects,and it also shows that there is a deep relationship between part of speech and named entities.This paper presents the improved NER method of DN with POS,which provides a reference for the promotion of NLP research in EMR,as well as positive guidance for medical information extraction,medical decision-making,and adjuvant therapy.