Research On Named Entity Recognition Methods For Clinical Medicine

Clinical named entity recognition is a fundamental step in many clinical natural language processing systems,aimed at identifying and classifying entity information such as diseases,symptoms,examinations,body parts,and treatments in clinical texts.Identifying useful named entities plays a crucial role in medical information processing and contributes to the development of research in the medical field.In recent years,with the development of deep learning technology,deep neural networks have been widely used in the task of Chinese clinical named entity recognition.Deep learning methods have achieved good results in open-domain named entity recognition,but the situation in the medical field is more complicated.Although there is a connection between the contexts in clinical medical literature and medical records,they have not fully utilized the global context between texts.information;on the other hand,in clinical medical texts,the nested structure of medical terms is complex,which makes it difficult for open-domain named entity recognition models to be directly applied to the clinical medical field.In view of the above problems,the content and innovations of this thesis are as follows:1.Aiming at the problem of language ambiguity in electronic medical records in the field of clinical medicine,a named entity recognition method for clinical medical texts based on pre-trained language models is proposed.This method is mainly composed of three parts: data embedding layer,feature extraction layer,and decoding layer.First,the data embedding layer uses the pre-trained language model ALBERT to learn to encode medical entity information and obtain the corresponding word vector representation;secondly,the feature extraction layer uses the Bi GRU model to capture contextual information to fully extract medical text information;finally,the decoding layer Partially employs conditional random fields to decode predictive medical entities.Applying the proposed method to the CMe EE dataset,the F1 value of 62.3% is obtained,which is15.3%,5%,3.5% and 1.8% higher than the baseline model,respectively,which verifies the effectiveness of the method.2.Aiming at the problem of nested entities in electronic medical records in the field of clinical medicine,a clinical medical named entity recognition method based on attention conditional random field is proposed.This method uses the ALBERT model in the data embedding layer,and still uses the Bi GRU network in the feature extraction layer.The decoding layer learns better entity representations by capturing the implicit distinctions and relationships between entities of different categories,allowing multihead attention mechanisms and conditional random field modules to perform residual connections,effectively alleviating the entity nesting problem in medical electronic medical records.The experimental results show that the proposed method is applied to the CMe EE dataset,and the F1 value of 71.1% is obtained,which is 11.6%,8.8%,7.3%and 5.4% higher than the baseline model,respectively,which verifies the effectiveness of the method.