Research On Medical Knowledge Graph Reasoning Algorithm Based On Graph Learning

Knowledge graph is a structured representation method of knowledge,which can find the relationship between various entities in the real world,and is widely used in different industry scenarios.In the field of clinical medicine,electronic medical record stores the real medical events of patients and contains rich clinical medical experience knowledge.By organizing massive electronic medical record data into medical knowledge graph,and using the information organization and reasoning technology of knowledge map,it has achieved obvious application effect in the direction of electronic medical record retrieval and clinical assistant decisionmaking.It is a meaningful and complex task to infer on the medical knowledge graph based on EMR,and make full use of the known clinical information and the internal relationship between them to predict the clinical decisions such as drug prescription and disease diagnosis.The existing methods still have a large space to improve the mining and utilization of EMR data.In recent years,graph learning related methods have shown great potential in knowledge reasoning,and have made new breakthroughs in network data analysis,intelligent recommendation,natural language processing and combinatorial optimization on graphs.Firstly,this paper constructs the medical knowledge graph based on EMR,and uses the graph learning method based on graph neural network to realize the information representation of medical knowledge map and the reasoning prediction of medication prescription.The main contents and innovations of this paper are as follows:(1)A time-series reasoning algorithm of medical knowledge graph based on co-occurrence statistical information is proposed:by transforming electronic medical record data into knowledge graph,modeling the structural association and time-series dependence between nodes of medical knowledge graph at the same time,and reasoning comprehensively considering the patient’s condition,the experiment shows that the model is more effective in large-scale real data set of electronic medical record The method is more effective in medication prediction.(2)A time-series reasoning algorithm of medical knowledge graph based on heterogeneous graph neural network is proposed:on the basis of the proposed algorithm framework,more abundant auxiliary data such as patient personalized information and laboratory test data are introduced,focusing on the heterogeneous characteristics of clinical event categories,the heterogeneous graph neural network is introduced,and based on this characteristic,a multi-layer attention mechanism is proposed to model the language between heterogeneous nodes semantic relevance.Experiments show that this method has achieved good results in drug reasoning task,and case analysis also verifies that fully considering the heterogeneity of medical events is conducive to better refining and representation of electronic medical records,so as to achieve more reasonable reasoning effect.