Research And Implementation Of Chinese Named Entity Recognition Algorithm For Medical Field

Electronic medical information system has accumulated massive medical data for medical research and medical practice.How to use natural language processing technology to mine potentially useful knowledge in medical data to improve modern medical level has become an urgent problem to be solved.Medical named entity recognition technology is the basis of mining medical text information.Based on the above background,this thesis focuses on the following three aspects:1.A BERT-PCA-BiLSTM-CRF model for medical entity recognition was proposed.First of all,the large-scale Chinese Wikipedia text and the small-scale medical text are used as corpus to train the word vectors,then the above word vectors are fused,and entity labeling results are obtained through bidirectional short and long-term memory network layer and conditional random field training.The experiment is compared with the current model based on deep learning,the results show that the model integrating general domain knowledge effectively solves the problem of insufficient feature extraction of complex general language in medical text sentences,and has better recognition ability.2.A BiLSTM-GCN-CRF model for medical entity recognition was proposed,which integrates the features of character space relations.First of all,the graph convolution neural network is used to learn the overall semantic features of the topological graph space of Chinese characters in the medical text,then the context dependent feature of medical text is extracted by using bidirectional long short memory network,then the muti head self-attention mechanism is introduced to fuse the above features,finally,the fusion feature to make use of condition random field decoding,get the best coding sequence.It is compared with the current algorithm model based on deep learning,the results show that the medical entity recognition model integrating the features of character spatial relationships solves the problem of insufficient semantic feature extraction and has better recognition ability.3.A medical named entity recognition system is designed and implemented.The system can directly call the trained medical named entity recognition model to recognize medical texts.At the same time,the system also includes electronic medical record management,entity labeling and other extended functions.