The Research Of Question Answering System Based On Knowledge Grap

With the continuous progress and development of society,healthy eating has become the focus of people's daily life.Together with the development of Internet technology,people can obtain recipe-related information more and more conveniently.However,without the thinking of the ”brain” of the machine,the answer to the question that is simply aggregated and disorderly can no longer satisfy people's demand.To meet people's demand for accuracy of information,people look forward to obtaining information about healthy diet in a more convenient and fast way through the intelligent question-andanswer system.Therefore,the question and answer system based on the recipe knowledge graph in this article is to use the recipe knowledge graph as the ”brain” of the recipe question and answer system,making the machine more ”cultural”.After the thinking of the ”brain”,answers with semantic understanding and a high degree of accuracy and strong interpretability are provided.After analysis,it is found that the current mainstream knowledge graphs are difficult to provide effective knowledge assistance to the intelligent question-and-answer system in the specific field of healthy eating,and there is very little research on the healthy eating question-and-answer system in the field of Chinese healthy eating.Therefore,in response to the above problems,this thesis has made a bold attempt to solve people's needs for healthy diet.The main work of this thesis includes the following three parts:This thesis first uses crawler scripts to crawl a large number of recipes,ingredients,recipes and medical effects of ingredients,as well as people who are suitable for recipes and ingredients from websites related to Chinese cuisine.Then in order to further enrich the relevant data in the recipes,this article obtained important data such as the nutritional value,scientific classification,and description of the food through Wikipedia based on the crawled food data.In the end,a large amount of data related to Chinese healthy eating with a high degree of completeness and wide coverage was obtained,filling the gap in the Chinese recipe field where there is no data related to healthy eating.Based on rule-based methods and dependency-based syntax analysis methods,this thesis extracts a large amount of prior knowledge of recipes related to healthy eating from semi-structured and unstructured data.After preprocessing the recipe data,this thesis extracts the concepts,terms and relationships related to the recipe field,and finally uses the Protégé ontology editor tool to generate the ontology in the Chinese recipe field.Under the guidance of the recipe domain ontology designed in this thesis,a new set of recipe knowledge graphs related to healthy eating is creatively constructed,and certain methods were taken to store the recipe knowledge graphs in the Neo4 j graph database,which solved the traditional knowledge.The map cannot provide effective prior knowledge for question-and-answer tasks in the recipe field.Finally,in order to apply the recipe knowledge graph to the question and answer task and in the case of fully considering the complexity of the relationship between the recipes,ingredients,population,efficacy and other data in the recipe data,two difficult question and answer tasks: simple tasks and complex tasks,are designed for the recipe question and answer task.Besides,high-quality question-and-answer pair was generated,and a data set for the recipe question-and-answer task was constructed.At last,a model based on graph convolutional network and pre-training model for recipe knowledge graph questionand-answer task is proposed,and after experimental analysis,it is proved that the model designed in this thesis can more effectively use prior knowledge to complete question-andanswer tasks than other models.It also shows that the recipe knowledge graph constructed in this thesis can provide effective support for recipe question and answer tasks.