The Design And Implementation Of The Question Answering System Based On The Knowledge Graph For Chinese Books

Most of the popular search engines still use the keyword matching to search and display information,however they don't use the knowledge to condense and give feedback.In recent years,knowledge graph has been claimed and widely used in the Question Answering systems due to its excellent performance in support of knowledge representation.By combining knowledge graph and Question Answering system it is likely to complete knowledge interaction with high quality.The CSSCI(Institute for Chinese Social Sciences Research and Assessment)of Nanjing University has a large collection of published Chinese book data without tools for management and query.Therefore,it is necessary and feasible to build a Question Answering system based on the knowledge graph for Chinese books(CBQA).Based on the analysis of knowledge graph and the related algorithms of machine learning,this thesis adopts the ELECTRA model and knowledge graph to build a Chinese book Question Answering system.The main tasks include:Firstly,a natural language problem classification model based on semantic understanding has been built.Based on the powerful understanding of semantics of ELECTRA and GRU natural language models,the system builds up a natural language model.Using existing knowledge in the book domain,a total of 400 questions which are divided in 22 categories are preset as a dataset,so as to realize the understanding of natural language problems in existing relationships.At present,the accuracy of natural language models for predicting problems is above 98% and the accuracy of entity recognition is above 87.9%.Secondly,a knowledge graph in the domain of Chinese books has also been built.This graph focuses on the field of Chinese book and it includes nine kinds of entities,ten kinds of relationships,a total of more than 700 thousand entities,and 3 million specific relationships.It completes a series of graph construction work such as data acquisition,information acquisition,knowledge fusion and knowledge processing.Finally,the back-end of this system has been successfully designed and implemented.It provides main functions such as natural language question answering,data management,user question completion and evaluation collection etc.For now,the system has been successfully deployed and implemented online,processing more than 295000 book data,and it can provide query and management methods for Chinese books.