Research On Chinese Medical Text Named Entity Recognition Based On Semi-Supervised Multi-Feature Model

The named entities covered in the Chinese medical text records contain a large amount of medical information that is closely related to the patient’s health.The rapid and accurate identification of the medical named entities in the Chinese medical text data is a key issue to promote the development of medical intelligence.However,the Chinese medical text records contain the patient’s personal information,so there is very little public Chinese medical text data,and the Chinese medical text data with annotation information is difficult to obtain,which seriously hinders the development of Chinese medical Named Entity Recognition.With the development of NER technology,using deep learning to carry out NER tasks has become the focus of researchers.The BiLSTM(Bi-directional Long Short-Term Memory,BiLSTM)model uses the context features in the extracted text data to achieve the purpose of identifying named entities.However,the text features extracted by BiLSTM are single,and the training speed of the model is slow.In this paper,for the problem of "single feature extraction and slow training speed",a mixed model of BiLSTM and IDCNN(Iterated Dilated Convolution Neural Networks,IDCNN)is proposed,the contextual features of the text and the surrounding features of the entities are extracted in parallel to realize the recognition of named entities in the Chinese medical text data,shorten the training time of the model,and improve the recognition effect of the model.In the recognition process,this paper also addresses the problem of "invalid tags" by adding CRF(Conditional Random Field,CRF)on the basis of the hybrid model to learn the constraint rules between tags to reduce the probability of invalid tags appearing in the recognition results.In addition,for the problem of "limited medical text with labeled information",this paper uses a semi-supervised learning method,combining medical text data with and without labeled information to train a multi-feature model.The feasibility and effectiveness of the semi-supervised multi-feature model NER method are demonstrated through experiments.In this paper,use three common news corpora of public data sets,combined with supervised and semi-supervised learning methods to train the multi-feature model.The experimental results verify the rationality and versatility of the proposed method.The multi-feature model proposed in this paper achieves the purpose of accelerating model training and improving the effect of NER,and the use of semi-supervised learning reduces the resource consumption of labeled text data,it is of great scientific significance and use value to carry out the task of Named Entity Recognition in the field with less labeled data.