Research And Design Of Reasoning And Question Answering System Based On Exploration And Development Knowledge Graph

The working outcomes in the field of exploration and development usually presented in the form of unstructured documents,such as papers and reports.However,the knowledge graph is to purify knowledge from the field.When knowledge is needed in field work,the full-text search and traditional question-answering system cannot obtain complex knowledge from the knowledge base,due to the level of the superficial keywords and semantic matching.To overcome shortcomings of the current systems,a generative method and a relation reasoning method,based on knowledge graph,are proposed to generate complex answers and to reason implication relation,respectively.For implementing the two works,the thesis designs a reasoning and question-answering system based on the knowledge graph of exploration and development.It mainly includes the following.(1)Design and implement an integrated question-answering system.It includes functions that are able to explore the methods of searching and querying into documents and knowledge base.Moreover,some related algorithms of retrieval have been optimized and improved on this basis to improve the system’s recall ratio.(2)Propose a generative model for the question answering based on multiparagraphs.It deals with the problems that answers distributing in multiple paragraphs.The model integrates knowledge graph information into the document representation to enhance semantic information,and generates the final answer with the pointer generation network.The model can match the answer content from multiple relevant paragraphs and summarize to generate a comprehensive answer.(3)Propose a question-answering model for cross-paragraph reasoning to deal with problems involving implicit relationships.The model constructs relations related to question in the documents and knowledge graph into a unified reasoning graph.Based on the reasoning graph,the relevance between is built up.Then the model expresses and calculates the reasoning graph with the graph neural network to get the final answer.The thesis designs and implements a reasoning and question-answering system based on exploration and development knowledge graph,which has a nice accuracy rate in several publicity data sets and exploration and development data sets.And the system has an application value for actual exploration and development work.