Chinese Medical Entity Recognition Algorithm Based On Fusion Feature Albert

The electronic medical record comes from the hospital information management system.It records detailed health information of patients and contains rich and valuable clinical knowledge.However,the current electronic medical records are mostly unstructured or semi-structured texts,and computers cannot directly use them for analysis.Biomedical named entity recognition technology is the basis for mining medical text information,which can identify specific medical entities such as disease symptoms and treatment from narrative medical texts.It provides data support for clinical decisionmaking,intelligent medical care,and hospital information construction.This thesis mainly studies the following two aspects:(1)Firstly,this thesis proposes a Chinese medical named entity recognition algorithm that combines Albert’s training features at different levels.According to the characteristics of the language structure of medical texts,this thesis optimizes the word embedding learning stage so that it can extract the word vector features of phrases,linguistics,semantics and so on in the medical text.Secondly,according to the existing labeling norms and the guidance of clinical experts,based on the BIO labeling system,this thesis constructs three real corpora of auxiliary examination,physical examination,and chief complaint.Then use the self-built corpus to test the feasibility of the model.The experimental results show that the F1 value of the fusion feature vector model proposed in this thesis reaches 0.9648,0.9630,0.8781 on the three corpora of auxiliary examination,physical examination,and chief complaint.Compared with the unfused algorithm model,it is 1.6%,0.43%,2.03% higher,which proves that the fusion feature vector model proposed in this thesis is feasible and effective.In addition,it is compared with the classic entity recognition model Bi LSTM+CRF and Bert+Bi LSTM+CRF.The experimental results show that the F1 value of the algorithm proposed in this thesis is higher than the other two algorithms,which proves that the fusion feature vector model proposed in this thesis has greater advantages in the task of Chinese medical named entity recognition.(2)Based on the research of(1),this thesis has completed an electronic medical record 360 view system.The system can not only store and manage electronic medical records,but also combine named entity recognition and regularization to display electronic medical records in a structured manner.Users can view 360 views of medical records in different display areas.For the convenience of users,the normal value display and search functions have been added to the system.The system also supports online annotated data for authorized users to build a corpus.The research of Chinese medical named entities is helpful to the prevention and control of diseases.The clinical knowledge mined by entity recognition technology provides data support for clinical decision-making and smart medical care.The clinical decision-making system assists medical staff in disease diagnosis and improves the quality of medical care.Smart Medical provides personalized treatment plans based on patient information.