Research On Construction And Application Of Medical Knowledge Graph For Stroke

The electronic development of clinical diagnosis and treatment has enabled massive clinical data to be stored on the Internet.For the problem of insufficient medical data utilization,the use of knowledge graphs to deal with loosely structured medical knowledge has a good effect.The thesis mainly studies the construction method of knowledge graph for the medical field and the application of stroke treatment,focusing on the construction of the medical field knowledge graph and applies the stroke knowledge graph to the stroke treatment mini program.The main content of the thesis includes:(1)Aiming at the problem of insufficient utilization of medical data,this paper proposed a method of constructing a knowledge graph in the medical field by sorting and summarizing it.It consists of four steps: firstly construct the ontology structure of the knowledge graph as the framework of the knowledge graph;then use Crowdsourcing semantic annotation to get Labeled regular data;then complete the external data,and integrate the wellstructured data from other sources with the semantically labeled data;finally,use deep learning algorithms to extract and supplement the loose entities or relationships in the knowledge graph(Named as MCDE method).The thesis conducts experimental analysis on the three steps of semantic annotation,entity set expansion and relationship extraction.Among them,the F1 value of semantic annotation is more than 90%,the accuracy of the expanded entity set is more than 80%,and the effect of relationship extraction is not obvious.(2)Based on the MCDE method,a knowledge graph of stroke was constructed.First,use a semi-automatically labeled stroke disease dictionary,combined with international standard medical terms such as ICD-10 to build the model layer of the knowledge graph;then,crawl the public medical text information of the vertical field medical website and the Crowdsourcing website,and after data cleaning,get Knowledge fusion of the medical knowledge triples and stroke information extracted from the Chinese symptom database.This part uses an improved entity similarity calculation method to expand the map;finally,the established knowledge graph is continuously improved through iterative updates.The MCDE method was used to construct a knowledge graph of stroke with higher accuracy in a relatively short period of time,which verified the effectiveness of the method to construct a medical knowledge graph.The role of Trans series models in the process of embedding the knowledge graph was compared and analyzed,and finally the Trans D model with a higher F1 value was selected.(3)Based on the above work,improve the stroke treatment applet based on the stroke knowledge graph,and use the stroke knowledge graph as the data layer of the intelligent retrieval module.The program provides services such as knowledge retrieval,emergency maps,data management,and emergency calls.Provide patients with health management tools,combined with deep learning models,realize the process of applying knowledge graphs to intelligent retrieval,and solve the end-to-end problem of structured data to applications.