Research On Event Extraction Method Of TCM Clinical Records Based On Deep Learning

The information resources contained in the text of clinical medical records are extremely rich,but due to the limited degree of standardization,the information can not be effectively extracted and utilized.The patient’s past condition,diagnosis and treatment records are of great significance to assist doctors in making better medical decisions.The research task of this paper is to extract the events from the clinical medical records of traditional Chinese medicine,present the patient’s disease development in the event unit,and extract the occurrence time of each event.Event extraction technology is to extract the events that users are interested in from unstructured information and present them to users in a structured way.It is found that the current mainstream event extraction methods are not suitable for the research work of this paper,whether in the research task or research scope.In view of the above problems,this paper studies the task of event extraction of TCM clinical medical records based on deep learning sequence annotation method,and mainly completes two tasks.Firstly,an event extraction tag structure based on sequence annotation is designed,which can extract the event and event time simultaneously;Then,aiming at the special research task of this paper,a set of multivariate event extraction evaluation system is designed.The main work of this paper is as follows(1)Data set construction.Because of the particularity of the research task,this paper does not use the open dataset applied by traditional event extraction.Instead,it uses the diagnostic medical record text of New Coronavirus pneumonia in five TCM hospitals in Hubei as the data set,and carries out the rule matching algorithm and manual audit to deal with the data set,so as to improve the normalization of the dataset.(2)The design of label system and evaluation system.In this paper,a unique label system is designed,using event and time dual attribute label nested annotation,and good application results are achieved.At the same time,two sets of evaluation criteria are designed for the task of clinical medical record event extraction.Among them,six different evaluation criteria are constructed according to the degree of relaxation.(3)Research and application of event extraction method based on deep learning.This paper adopts the deep learning method of BiLSTM + CRF,combined with the tag system method designed in this paper,extracts the event and time from the text of TCM clinical medical records,and achieves good results.Then,the results of event extraction and entity extraction are combined and applied.Based on the location of the two,the time attribute of the entity is obtained,which provides data support for longitudinal clinical research.