Research On Automatic Question Answering And Question Generation Based On Deep Learning

Teaching machines to understand texts in the form of natural language is an elusive and long-term challenge for artificial intelligence.This thesis focuses on two major tasks in natural language processing: automatic question answering and question generation.Among them,automatic question answering means that the machine can answer questions about natural language forms within its knowledge scope when it has a certain knowledge reserve.Question generation means that the machine can ask questions about a text after reading and understanding it.On the one hand,this thesis believes that automatic question answering and question generation have gradually become a new trend of human-computer interaction.On the other hand,the research on these two aspects is extremely important for the development of other AI applications such as dialogue system.The main works and innovations of this thesis are as follows:(1)this thesis tackled the construction technology of knowledge graph;completed the construction of knowledge graph in the limited domain;designed and implemented the automatic question answering model based on knowledge graph.In view of the cases that would appear in question answering,such as the complex logic of questions and the triples related to question entities can not be found in the knowledge graph,the methods of decomposing multi-entity and multi-relation questions logically and constructing query statements by computing semantic similarity between knowledge triples and questions are pioneered respectively.And these two methods can improve the understanding ability and recall rate of question answering system.(2)the question generation technology based on knowledge graph is studied.Aiming at the disadvantage of the prior technology that can only question a single knowledge triple,and combining the advantages of the template-based question generation method,a question generation model based on multiple triples is built.For any given number of triples,the model can automatically generate question for the specified answer,which improves the diversity and difficulty of question generation system.(3)the question generation technology based on free text is studied.Using the principle of encoder and decoder,the question generation model based on free text is designed and implemented.According to the given free text,the model can automatically generate question with natural language form for the specified answer after understanding a document.