Construction Of Knowledge Graph For Stroke Electronic Medical Records Based On Deep Learning

[Purpose] The purpose of this article is to do the Named Entity Recognition(NER)and Relation Extraction in stroke Electronic Medical Records(EMR),thereby construct the stroke EMR knowledge graph,and lay the foundation for the upper-layer application of Artificial Intelligence(AI)in the medical field.[Methods](1)Literature analysis method: combing the research status of this topics at home and abroad,summing up the common methods of medical information extraction and the general process of knowledge graph construction,summarizing the classification standards of medical naming entities.(2)Expert interview: consult experts in medical fields to improve the classification standards of named entities,and formulate a detailed corpus labeling scheme,so as to continuously improve the data quality.(3)Deep learning: build a NER model based on Bi LSTM-CRF,and build a relationship extraction model based on Bi LSTM-Attention.(4)Statistical analysis method: the Precision,the Recall and the FMeasure are used to evaluate the effect of models,so as to repeatedly adjust and optimize models.[Results](1)The entity categories of stroke EMR to be identified are divided into 5 categories,and the relationship categories to be extracted are divided into 6 categories.(2)The Precision of Bi LSTM-CRF model is 91.59%,the Recall is 92.78%,and the F-Measure is 92.18%.The Precision of Bi LSTM-Attention model is 75.15%,the Recall is 68.50%,and the F-Measure is 68.64%.(3)After the unification of the entities,nodes and relationships were created in Neo4 j by triple data,and the stroke EMR knowledge graph was obtained.[Conclusions](1)Due to the structure and language characteristics of EMR,when to do NER and Relationship Extraction,we need to consider improving the quality of the corpus by constructing professional dictionaries and generating entity groups,so as to obtain a better model effect.(2)In the knowledge graph,node can be arbitrarily selected to obtain a semantic network centered on that node.A third node can be used to establish a connection between two nodes that do not have a direct association.