Research And Implementation Of Multi-Round Question Answering System Based On Medical Knowledge Graph

With the continuous development of society,people pay more attention to health problems.Under the "Internet+Healthcare",patients will search for disease-related information online before seeking medical treatment.However,patients who search for medical information online may encounter problems such as delayed information feedback and the need to filter information on their own.In this context,many scholars have conducted research on medical question answering system,which includes three stages:question parsing,answer retrieval and answer generation.In the practice of medical question-answering,two issues are worth noting.First,the intent types and expression forms of medical field questions are diverse,which leads to inaccurate intent parsing.Second,patients usually need to ask the same question multiple times to obtain the desired results.Due to the omission of some content in the question,the absence of entities or intent can cause difficulties in accurate answer retrieval,which brings challenges to the answer retrieval stage.This article carries out corresponding research based on these two prominent problems,as follows:1.This paper proposed a question parsin g method based on fusion model.This method uses the idea of parameter sharing to fuse entity recognition and intent recognition tasks.The two tasks are mutually learned by sharing vectors in the embedding layer,fusing vectors in the encoding layer,and updating network weights with the sum of loss functions.When TextCNN is added into intent classification,the accuracy of medical intent recognition is improved further by extracting context information of intent features using convolutional kernel sliding in sentences.Experimental results on real data sets show that the fusion model proposed by the paper improves the accuracy of traditional single medical intention recognition task and question parsing.2.This paper proposes a method of completing missing questions using local knowledge graph.Firstly,the types of missing questions in the medical question scenario are analyzed and summarized.Secondly,a completion solution is proposed for the missing questions.The knowledge graph stores entities,entity properties,and relationships between entities in the form of a graph structure,containing rich semantic knowledge.Therefore,it has the characteristics of interpretability and knowledge reasoning.As missing questions need to be semantically inferred through relationships between historical questions,the semantic knowledge of the knowledge graph can be applied to complete missing questions.Compared with classical methods for missing question completion,the local knowledge graph completion method shows better performance in completing the tail entities.3.This paper designs and implements the multi-round question answering system based on the medical knowledge graph.The question answering system is built using the Flask framework in Python,which integrates the question parsing fusion model and the question completion method proposed in this paper.The system mainly includes medical knowledge graph retrieval and multi-round question and answer function modules.