Question Answering System Based On Knowledge Graph Of Elementary Education

In the era of artificial intelligence,the internet search is evolving from the keywords search to the intelligent search.In intelligent search,knowledge graph is the basis while automatic question answering is the engine.They have played important roles in many industries.Elementary education is the cornerstone of national education,how to make full use of information means to provide richer teaching methods and after-school tutoring support for teachers and students is a kind of meaningful work.On the basis of knowledge graph,my thesis focuses on the difficulties and challenges of question answering systems,and finishes the following work:Proposed a hybrid question answering method based on knowledge graph and massive text data.Question answering by template matching method searches for knowledge graph,and all the answers came from the knowledge in knowledge graph.They were accurate and of high quality but with limited coverage.Question answering by information retrieval method matches the text which relates to the question,and took the text with the highest confidence as the answer,this method has wide coverage,but the accuracy is low.The system absorbed the advantages of the above method,to ensure the system's performance in accuracy and breadth.Constructed a question answering system that based on knowledge graph.The system's data source is the knowledge graph of elementary education and a mass of internet text.It focuses on template matching method and takes information retrieval method as a supplement.The purpose is to ensure system performance with precision and coverage.Constructed a framework which could extract the entity's attribute value and generate new triples automatically.The completeness of knowledge graph has a direct influence on the performance of the question answering system.To enrich the data of knowledge graph,this thesis focuses on the existing entity in knowledge graph that lacked attribute value,and used “transfer learning” method,this thesis applies the solution of machine reading comprehension to knowledge graph's attribute value supplement task.