Research On Intelligent Diagnosis Integrated With Knowledge Graph

As the people’s demand for medical services grows,the imbalance between supply and demand in China’s medical and health services has become increasingly prominent.The promotion of medical informatization has enabled major medical institutions to accumulate massive amounts of electronic medical records,which is the big data in the medical field.The knowledge graph has developed rapidly in recent years and has become the infrastructure of knowledge-driven intelligent applications.Based on the electronic medical record big data and knowledge graph,the combination of datadriven and knowledge-driven will help doctors make decisions and improve work efficiency.It is also an important way to alleviate the contradiction between supply and demand of medical services.In this thesis,the intelligent diagnosis task is treated as a multi-label classification task,and mainly discusses the effective method of fusing the knowledge graph into the intelligent diagnosis task.The specific research results are as follows:(1)A Knowledge powered Attention and Information-Enhanced(KAIE)model is proposed for intelligent diagnosis research on fusion of triplet information.The numerical information and key information in electronic medical records play a more important role in diagnosis.The KAIE model differentiates key information from other texts by differentially processing key information in the input part.The introduction of numerical information uses a multi-head attention mechanism for processing.For the triples obtained from the entity linking method,the knowledge graph embedding method and multi-way attention mechanism are proposed to fuse the triple information with the deep learning model.Experimental results show that,compared with traditional machine learning and commonly used deep learning models,the KAIE model can effectively improve the performance of intelligent diagnosis due to its information enhancement and fusion of knowledge graph triplet information.Compared with the BERT model,its F1 value has increased by 1.37% to 81.11%.(2)A Graph-based Structural Knowledge-aware Network(GSKN)model is proposed to conduct intelligent diagnosis research on fusion graph structure information.In addition to the shallow triplet information in the knowledge graph,the triples also contain complex relationships,and these relationships can be expressed as the type of graph structure information.On the basis of the KAIE model,the GSKN model constructs triples into the form of knowledge graph subgraphs,and uses GCN and interactive attention mechanism to encode and merge them.Experiments show that the diagnostic effect of the GSKN model with the introduction of graph structure information has been further improved compared with the KAIE model,the F1 value reached 81.42%.(3)Based on the above-mentioned intelligent diagnosis algorithm,an intelligent diagnosis platform including medical record management,medical record quality control,intelligent diagnosis and other functions has been developed and constructed.This platform can provide doctors with medical record quality control and clinical decision-making assistance.