Research And Development Of Multi-round Dialogue Technology For Service Robots Based On Medical Knowledge Grap

With the improvement of living standards,people are paying more attention to their health.However,the problem of insufficient and uneven development of medical resources still exists in China.Applying computer technology to medical conversations to provide more efficient and accurate health consultation solutions has become a hot topic of concern among scholars at home and abroad.Although single-round retrieval-based questionanswering systems based on knowledge graphs have become relatively mature,there is still little exploration of multi-round conversations.In this context,this paper designs and develops a multi-round conversation system for service robots based on medical knowledge graphs,providing new directions and technologies for the development of medical service robots and reducing the burden of medical inquiries on doctors.The main research contents are as follows:1)In order to improve the accuracy of natural language understanding in Chinese medical dialogues,this paper collected commonly asked patient health consultation questions in the Chinese medical dialogue domain,designed and annotated the Chinese Medical Information Seeking Dataset(CMISD-UQS),and proposed a medical consultation intent understanding and entity extraction algorithm based on an intent-slot attention mechanism.This algorithm explicitly establishes the correlation between intent and slot to reduce error propagation while improving the interpretability of the algorithm.On a public dataset,the joint recognition effect of this paper’s algorithm is superior to 8 other benchmark algorithms,and the minimum variance of sentence-level semantic frame accuracy is 0.02,indicating that the stability of the algorithm is strong.On the CMISDUQS dataset,the medical intent recognition accuracy,semantic slot filling F1 value,and sentence-level semantic frame accuracy of this paper’s algorithm are 78.1%,94.9%,and73.2%,respectively,all outperforming other comparative algorithms.2)In order to improve the modularity and scalability of the dialogue system and make the system respond quickly to user requests,a pipeline approach was chosen to develop the medical dialogue system.First,a knowledge graph for daily medical and health consultations was built as the knowledge data base.Secondly,the proposed intention understanding and entity extraction algorithm for medical consultation based on intentionslot attention mechanism is integrated into the natural language understanding module of the dialogue system to improve the accuracy of the system in understanding user input.In addition,a support vector machine classification model is selected for user intention determination,and ChatGPT’s API is integrated into the dialogue system as part of the user gossip module.Finally,the Flask framework was used to build the web page of the system and the front and back-end interaction with the Neo4j database.3)The algorithm model was deployed to the home service robot platform,and a multiround dialogue system based on medical knowledge graph for service robots was implemented with Python language integration based on the given system architecture design,process planning,and database construction.