Research And Application Of Key Techniques For Named Entity Recognition Of Electronic Medical Records Based On Deep Neural Network

With the development of "Internet+Medical",more and more researchers are turning their attention to the use of computer technology to extract medical information,especially the new crown pneumonia that broke out in late 2019,which has attracted the public’s attention to the medical field..Chinese electronic medical record named entity recognition is the first step to extract medical information from electronic medical records,which has great research significance and value.Based on the research on the existing Chinese electronic medical record naming entity technology,this paper proposes a multifeature fusion character-level word embedding algorithm,and on this basis,proposes a WC-LSTM Chinese based on self-attention mechanism.Finally,the Chinese electronic medical record named entity recognition system is realized based on the above model.The research contents of this article are:(1)This thsis studies the characteristics and difficulties of character-level word embedding algorithms for Chinese electronic medical record named entity recognition from the aspects of language characteristics and application fields.Based on the Skipgram algorithm,character-level word embedding algorithm of radical radical information and pinyin is proposed.The algorithm is valued by its external evaluation.(2)This thsis proposes a WC-LSTM Chinese electronic medical record named entity recognition model based on multi-head self-attention mechanism.Aiming at the problem of missing word information based on the character-based named entity recognition model,this article uses the method of incorporating the information of the shortest word corresponding to the character into the word vector to supplement the word information.And,the self-attention mechanism is used to strengthen relationship of characters between which the distance are far away.Reducing the noise that may be caused by using the shortest word.At the same time,the structure of the model is stabilized by filling the word information in the word vector without the corresponding vocabulary with <PAD>,which can perform batch training,which greatly reduces the training facility.The feasibility and effectiveness of the model were verified through comparative experiments.(3)Based on the research of Chinese electronic medical record named entity recognition model,a Chinese electronic medical record named entity recognition system has been implemented.This system has a simple authority management function,avoids unnecessary data leakage risks,and not only automates electronic medical records Named entity recognition,annotation,and results can be modified manually.