| The question answering system can obtain the required information quickly and accurately from the massive data.Technology is changing rapidly,and the revolutionary development of deep learning has brought great progress to question answering system and developed a series of natural language processing technologies.At the same time,the knowledge graph keeps rising and more and more knowledge is stored in it.The knowledge graph has become an important source of the answers of the question answer system,which is widely concerned by the industry personnel.Thesis focuses on the question answering system based on knowledge graph.In view of the disadvantages of traditional question answering system,thesis designs and implements a multi-turn question answering system based on the knowledge graph of computer literature domain.The purpose of the thesis is to provide a user-friendly question answering system to help scholars in the field master the basic information of computer-related literature and facilitate computer-aided instruction.The main work contents are as follows:(1)Entity alignment of knowledge graph in computer literature domain is realized.Considering the extensive knowledge in the existing computer literature field,there is no unified structured data.Sci KG is used as the basic knowledge graph,entities in DBLP dataset are aligned with it and entity attributes are fused by similarity judgment.A knowledge graph with rich content and more extensive content is obtained.This provides basic data support for the later construction of question answering system.(2)Based on the improved Long Short-Term Memory unit structure,the joint model of intent detection and slot filling is established.In the thesis,the multi-task joint training modeling method based on the improved neural network is adopted to avoid the error amplification of the cascade task,optimize the question comprehension task and save the time of model training.Multi-group comparative experiments were carried out on datasets ATIS,Snips and the computer literature domain question dataset CSLQ constructed in the thesis.The experimental results of the constructed model are superior to the other models mentioned in the thesis on the three datasets.Compared with the baseline model in CSQL,the F1 value of slot filling task is increased by 5.46%,and the accuracy of intent detection task is increased by 4%,the intent accuracy of semantic is increased by 6.19%.(3)The rules of multi-turn question answering mechanism are formulated,and a domain question answering system based on knowledge graph is designed and implemented.Combined with the slot filling task,the slot template is designed and the slot inheritance mechanism is formulated.When the user does not make clear expressions,it can ask the user and interact,guide the user to ask questions,and improve the user’s experience.Finally,the design and implementation of the multi-turn question answering system based on the knowledge graph of the computer literature domain is completed,and the system is tested. |
