Research On The Construction And Application Of Knowledge Graph For Intelligent Triage

With the continuous growth of the amount of medical data,the use of data mining technology to mine valuable knowledge from complex medical data has become a hot topic.At present,the information carrying capacity of medical data is large,and the structure is relatively loose.Therefore,the construction of large-scale knowledge graphs has become the most basic and critical task,and the level of its quality can directly affect the effect of the application.Now for the task of intelligent triage,it is basically traditional data-driven triage,and more is to classify symptoms from the dimensions of symptoms and diseases to achieve the purpose of department recommendation.However,as the core medical data of diseases,symptoms,drugs,departments,etc.is increasing,multi-dimensional data references are provided for intelligent triage.Therefore,it is necessary to integrate multi-dimensional medical information.Combining the strong knowledge,professionalism and explanatory characteristics of the knowledge map,this article considers introducing the knowledge map into the field of triage to promote the accuracy of triage and ensure the interpretability of triage results.The main research contents of this paper:(1)Based on the BERT-Bi GRU-CRF entity and relationship joint extraction method.In order to solve the problems of error propagation in the existing pipelined extraction process,failure to consider the implicit relationship between entities,and long model training time,this method takes into account the structure of the model,context and semantic information,first uses BERT to vectorize the text,and then passes through two gates The Bi GRU network performs forward and backward semantic learning,and finally outputs the optimal label sequence through the CRF layer.Finally,it is verified that the F1 value of the proposed model is better than the LSTM method,and 200 rounds of model training can save about 3 hours,and the overall model training efficiency and effect are better.The model can reveal the relationship between diseases,symptoms,onset locations,and various entities in departments,and provide a relatively high-quality data basis for intelligent triage.(2)This paper proposes a BERT entity alignment method,which takes into account the attributes and semantic characteristics of entities.And this method increases the accuracy of entity alignment by learning the semantics and semantic information of entities,which are used as the screening criteria for entity alignment.Through this model,standard entity pairs can be aligned,and then the triples can be stored in Neo4 j to complete the construction of high-quality large-scale knowledge graphs.(3)Knowledge reasoning method based on Trans D.considering the complex semantic relationship,the Trans D method is proposed to represent the many-to-many relationship in the knowledge graph.Then the patient complains Data is extracted for key entities,and the extracted entities are linked to the triple vector by using the largest common subsequence method to form a standard entity set,which is used as an input set to calculate the similarity through knowledge inference.Finally return the department list information with high similarity.(4)An intelligent triage system based on knowledge graph.According to the needs of patients and triage nurses,the main process and user process of the intelligent triage system are analyzed,then the structure and functions of the triage system are designed,and finally the intelligent triage system is realized.