Construction Of Knowledge Map For Type 2 Diabetes Based On Deep Learning

Type 2 diabetes is one of the most important categories of diabetes.Among all people with diabetes,type 2 diabetes accounts for more than 90%.By 2021,China has nearly 140 million diabetes patients,which is the country with the largest number of diabetes patients in the world.Most of them are type 2 diabetes patients,especially the elderly,and the age of the affected population is getting younger and younger.Type 2diabetes can also cause a variety of complications.The mortality and disability rate caused by complications are very high.In 2021,about 6.7 million people died of diabetes and its complications in the world,equivalent to one diabetes patient dying every five seconds.At present,the treatment of type 2 diabetes patients mainly depends on self prevention and treatment.Therefore,it is particularly important to publicize,educate and make rational use of the knowledge of prevention and treatment of type 2 diabetes.Based on the open source data set of diabetes provided by Ruijin Hospital,this paper constructs the medical knowledge map of type 2 diabetes by using the method of deep learning.The knowledge map of type 2 diabetes constructed in this paper can extract the relevant knowledge of type 2 diabetes from professional medical literature or other relevant medical texts,store the extracted knowledge,obtain the semantic knowledge base containing the relevant knowledge of type 2 diabetes prevention and treatment,and finally realize the intelligent analysis and utilization of the knowledge of type 2diabetes prevention and treatment.This paper constructs the knowledge map of type 2 diabetes through three main tasks: named entity recognition,entity relationship extraction and knowledge storage.In this paper,the named entity model based on ELECTRA-CRF model is constructed for the first time in the named entity recognition task of the knowledge map of type 2diabetes.15 types of entities such as disease name,detection method,treatment method and drug name in the diabetes data set of Ruijin Hospital are automatically recognized.The accuracy,recall and F1 value of entity recognition are taken as the model evaluation criteria,and are compared with bert-crf model and bilstm-crf model respectively,The results show that the ELECTRA-CRF model constructed in this paper has achieved better entity recognition effect,and ELECTRA-CRF model can achieve better training effect in a shorter time.In the task of entity relationship extraction,this paper constructs an AGGCN model based on attention mechanism for entity relationship extraction,and compares the experimental results with pa-lstm model,cgcn model and sdp-lstm model.The experimental results show that the AGGCN model constructed in this paper has better entity relationship extraction effect in type 2diabetes data set.In the task of knowledge storage,this paper uses the neo4 j graph database to store the extracted entities and entity relationships,and completes the construction of the knowledge map of type 2 diabetes.Finally,the constructed knowledge map of type 2 diabetes can further provide reliable data support for the medical system in different scenarios.It has positive significance for the diagnosis,prevention,medication guidance and rehabilitation treatment of patients with type 2 diabetes.