Medical Text Named Entity Recognition Based On Improved Sequence Labeling Model

With the development of neural network models,the combination of neural networks and traditional sequence labeling models has gradually replaced the traditional sequence labeling models based on hidden Markov or conditional random fields,and has become the mainstream direction in the field of named entity recognition.Medical texts come from medical professional books and are important materials for constructing medical knowledge graph.Medical text named entity recognition plays an important role in the construction of medical knowledge graph.Based on this background,we have carried out series of researches on the method of the named entity recognition on medical text.First of all,we construct a medical text named entity recognition model that incorporates multi-granularity text features.Because multi-granular text features can solve the problem of the words outside the dictionary and enrich the representation of words,we construct a named entity recognition model that fuses character,subword,and word level features based on BiLSTM(Bi-directional Long Short-Term Memory)and CRF(Conditional Random Field)model.And then we compare the effects of context-free GloVe(Global Vectors)word vectors and context-related BERT(Bidirectional Encoder Representation from Transformers)word vectors on named entity recognition.Experiments show that the context-sensitive word vectors and the fusion of multi-granular text features can improve the effectiveness of named entity recognition.Secondly,for the problem of long sequence modeling,we build a medical text named entity recognition model based on the multi-head attention mechanism.Because the LSTM(Long Short-Term Memory)is prone to lose the long-term dependency of sequence content,we use a Transformer encoder based on the multi-head attention mechanism to replace the BiLSTM for sequence modeling.Experiments show that the Transformer encoder can improve the effectiveness of named entity recognition.Finally,we carry out the research of the problem of medical text named entity recognition on the condition of adding additional information.In this research,we add disease name and entity type as additional information based on the characteristics of the medical text,and finally the problem was converted into a reading comprehension problem.Experiments show that adding additional information can improve the effectiveness of named entity recognition,and the addition of entity type information has the greatest effect on named entity recognition.