Design Of A Clinical Medical Question Answering System Based On Knowledge Graph And Deep Learning

With the development of Internet technology and the evolution of artificial intelligence technology,the question answering system based on search engines cannot meet the increasing needs of people.People need to find the information they need in the complicated and lengthy search results.Under the influence of the "COVID-19" epidemic,how to accurately and simply understand the information of the illness has become an urgent issue to be solved.The clinical medicine intelligent question answering system implemented in this thesis can understand user questions,accurately identify user intentions,find answers in the knowledge base,and return the answers to users accurately and concisely.This thesis crawls the medical knowledge web resources on the Internet,and formalizes the extracted resources based on rules,and structurally generates the entity attributes and relationship attributes required by the knowledge graph,thereby constructing the knowledge graph of the corresponding field.Then this thesis use the open source Neo4 j graph database for organization,storage and query.Aiming at the user’s question intention and the relational linking algorithm,this thesis uses the relational linking algorithm of the combined model structure to calculate the corresponding correlation degree between the question intention and the relational attributes.Combine the query sentences of the knowledge graph to realize the algorithm layer of the question answering system.For the process of constructing the knowledge graph,this thesis conducted experiments on the web crawling technology and the technique of identifying entities and relationship attributes based on the rule base.The final construction of the clinical medicine knowledge graph has an entity scale of 80,000 orders of magnitude,and a relationship scale of 1450 thousands orders of magnitude.For the realization of question answering algorithm based on deep learning,this thesis analyzes the combined model effect through comparative experiments and proves that the combined model can take into account the advantages of convolutional neural network and recurrent neural network,so as to achieve better results.At the end of this thesis,through comparative experiments,it is found that the effect of adding the attention mechanism is improved,which proves the effectiveness of the attention mechanism for improving the effect of these several neural network models.Therefore,the question relationship attribute link algorithm of the CNN-Bi-LSTM-Attention combined model is the best.This thesis finally uses this combined model algorithm to implement the algorithm layer of the question answering system.Based on the above-mentioned theories and algorithms,this thesis constructs the structure of the question answering system,and implements a question answering system with a visual interface for clinical medical knowledge graphs.Under the experiments in this thesis,the system has achieved a higher accuracy rate and better stability,which proves that the system has a higher availability.In future research,more methods of constructing knowledge graphs can be used.The method in this thesis can be transferred to more fields to build a question answering system in other fields.