Medical Knowledge MAP Construction Based On Chinese Language Processing And Deep Learning

The knowledge map shows the key points of knowledge and internal links in the information in a visual form.The map of medical knowledge is the cornerstone of realizing intelligent medical care and is of great significance for clinical decision support and personalized medical services.Knowledge maps have not been widely used in the medical field,mainly due to the difficulties of unstructured text extraction and Knowledge map drawing.This paper adopts deep learning method to extract information from Chinese electronic cases.The electronic medical record has a detailed record of the patient's medical activities,contains a large amount of medical knowledge related to the patient's health status.Bidirectional cyclic neural network technology is used to extract important knowledge from electronic medical records(EMR).It expands the concept of subject domain knowledge mapping to the medical field.Based on a deeper investigation of existing entity recognition and relation extraction methods,a framework system of medical knowledge map based on Chinese language processing and deep learning is constructed.Mainly completed the following work:(1)The in-depth study of the concept of deep learning and related technologies has been conducted.Deep learning is a subfield of machine learning and a hot topic in recent years.Deep learning is a multi-level non-linear learning structure that can approximate complex functions.Compared with traditional machine learning methods,it better simulates human brain learning and achieves good learning results.(2)The BiLSTM-CRF model was used to identify the entities in the electronic medical record,and the five named entity categories of body parts,symptoms and signs,examination and test,treatment,disease and diagnosis were identified.The accuracy rate was 96.29%,and the precision rate was 91.61%,The recall rate was 96.27% and the F value was 93.96.The evaluation indicators of the five entity categories are also relatively high.The highest is the entity category of symptoms and physical signs.The precision rate is 95.97%,the recall rate is 98.69%,and the F value is 97.31.The lowest is the treatment entity category,and the precision rate is 84.98%,the recall rate is 91.67%,and the F value is 86.55.(3)BiLSTM-ATT model is used to extract the relationship between identified five kinds of entities.it defines relationships between entity classes,including signs and symptoms of body parts,diseases and diagnoses occurring in body parts,diseases and diagnoses leading to symptoms and signs,examinations and tests(inspection methods)find disease and diagnosis,examinations and tests(inspection methods)found signs and symptoms,treatment was applied to the disease and diagnosis,take treatment after examination and test,treatment improved symptoms and signs,body parts showed examination and test(abnormal examination results),treatment applied to body parts class relations.Experimental results show that the BiLSTM-ATT network model can be well applied to the extraction of entity relationships in Chinese electronic medical records.(4)The identified named entity and the extracted entity relationship are imported into the Neo4 j graphic database,completing the task of constructing the knowledge map based on the Chinese electronic case.