Research And Application Of Diabetes Question-answering System Based On Knowledge Graph

In recent years,with the improvement of residents’ consumption level and the acceleration of population aging process,the number of patients with diabetes has increased rapidly,and the trend of youth is obvious.According to the survey,effective prevention and treatment of diabetes includes knowledge education,self-monitoring of blood sugar and healthy diet.Over the years,the "Internet +" medical education and popular science service has accumulated a large amount of digital diagnosis and treatment information.Therefore,this thesis mainly studies how to help users acquire the knowledge of diabetes prevention and treatment simply and effectively through the Internet.Through the traditional search engine to query knowledge,users get redundant data,which is difficult to obtain knowledge quickly and accurately.The question-answering system can feed back the answer according to the user’s intention through the semantic analysis of the user’s question sentence,and the knowledge graph can realize the integration of large-scale data,which can use query sentences to query effective data.Therefore,the Diabetes Questionnaire based on the Knowledge Card will be studied and implemented,and an in-depth model of recognition of these entities will be used to improve the system’s ability to parse questions.After testing,the system will be able to respond quickly and accurately to users’ knowledge of diabetes prevention and treatment,and also respond to national intelligent medical policy.The main contents of this thesis are as follows:（1）Building knowledge graph in the field of diabetes.Web crawler technology based on Python crawls information related to diabetes in medical websites such as seeking medical advice,and cleaning and processing knowledge.In order to improve the quality of the knowledge graph and the ability of knowledge expression,the data extracted from the knowledge is fused with the diakg knowledge graph data set to complete the construction and persistence of the knowledge graph in the field of diabetes.（2）Research on named entity recognition algorithm.First,collect the question-answering dataset from the Chinese Diabetes Community as the self built corpus of question entity recognition,and segment the data and entity annotation.Then,through the design,the entity recognition model based on BERT + Bi LSTM + CRF is constructed.Experiments were carried out on ccks-2017 data set and self built data set,and the results were compared,which verified that the recognition effect of this model was significantly improved compared with other models.（3）Build a complete diabetes question-answering system.Based on the above research results,the function of question-answering system is realized by the combination of template based and semantic parsing.The specific implementation process includes: Chinese word segmentation,problem classification,named entity recognition,question syntax dependency analysis,query generation and knowledge retrieval.The knowledge graph of diabetes has been constructed as the data base of question-answering system,using Flask framework and D3.js tool packages the system,requests services through browser access,and finally realizes the automatic question-answer function and knowledge graph visualization function.Through the above work,this thesis first completes a high-quality diabetes knowledge graph,then uses deep learning technology to improve the semantic parsing ability of the system to the user’s questions in the question-answering task,and finally realizes a diabetes questionanswering system that can effectively and conveniently solve the user’s question-answering needs.