Design And Implementation Of Knowledge Graph Question Answering System Based On Deep Learning

With the popularity of the Internet,the rapid growth of data volume has led to more and more information,and people's requirements for information accuracy are getting higher and higher.Traditional search engine based on keyword retrieval returns sorted and related documents.Users still need to manually retrieve answers.Different from traditional search engine,the question answering system receives natural language questions and returns simple and accurate answers,which can help users get the target information quickly and accurately.The question answering system over knowledge graph is an important branch of question answering system,which answers natural language questions according to structured knowledge of the knowledge graph.Access to knowledge graph requires specific query statements.Since natural language is difficult to directly correspond to structured query,the mapping becomes a difficulty in the research of knowledge graph question answering.Solutions based on rules and vocabulary mapping require a lot of manpower which have low coverage and lack of flexibility.Traditional machine learning solutions rely on the effectiveness of artificial features,of which the effect needs yet to be improved.In recent years,deep learning has developed rapidly,which has brought many breakthroughs in the field of natural language processing.The sequence-to-sequence model based on neural network has been proved to have significant effects in the sequence transformation task.This thesis relies on the sequence-to-sequence model of deep learning to map question into query statement.On this basis,this thesis designs and implements a knowledge graph question answering system based on deep learning.The system of this thesis is divided into Web service module,question preprocessing module,question understanding module,query statement generation module,answer generation module,system log and knowledge graph storage module.The question preprocessing module performs syntactic analysis and entity identification.The question understanding and query statement generation module improve the sequence-to-sequence model for mapping question into query statement.The answer generation module obtains the answer information by querying the knowledge graph according to the query statement.For the mapping process,this thesis starts with the syntactic structure and proposes a Hierarchical Attention Mechanism based Model for Question Answering over Knowledge Graph(KGQA-HAM),which is composed of encoding structure and decoding structure.The encoding structure encodes the subtrees of each layer of the question dependency tree,which establishs mapping relationship between the question and the query statement.The decoding structure extracts entity or relation semantics of the question based on the hierarchical attention mechanism,and integrates them into the neural network to generate query statements.This thesis implements comparative experiments.The experimental results show that the KGQA-HAM model proposed in this thesis significantly improves the accuracy of mapping the question to the query statement and the F1 value of the question answering system.Through system testing,the function and performance of the question answering system designed and implemented in this thesis both meet the expected results.