Design And Implementation Of Question Answering System Over Knowledge Graph Based On Reinforcement Learning With Curiosity Mechanism

As the information on the Internet becomes more and more complicated,it is an inevitable demand for every Internet user to quickly and accurately acquire the knowledge they need.Most of the traditional search engines use retrieval-based methods to give the relevant webpage information of keywords provided by users,but can not directly answer the users' questions.Question Answering system meets this requirement well,and to achieve this goal,the key is to accurately transform natural language statements into structured query statements that can be executed on computer systems.The research goal of this paper is to train a semantic parser through reinforcement learning.It can transform natural language problems into formal queries that can be executed on the Knowledge Graph,thus realizing a Question Answering system on the Knowledge Graph.The system mainly consists of two parts,one is entity linker and the other is sequence-to-sequence generator.The entity link in this paper uses the existing entity linker,and focuses on how to use the reinforcement learning training sequence generating model.The main contributions of this paper are as follows:(1)A Question Answering over Knowledge Graph method based on reinforcement learning is proposed.This method defines a series of actions in advance,and automatically trains how to generate formal queries corresponding to natural language questions by using natural language questions and corresponding answers without labeling formal queries.(2)A method of generating reinforcement learning sequence with curiosity mechanism is proposed,which uses curiosity mechanism to solve the problem of sparse reward in reinforcement learning,so as to encourage agents of reinforcement learning to explore unknown space and accelerate the training of reinforcement learning.Experiments are carried out on two general knowledge graph in Chinese and English and a self-defined military knowledge graph.The results show that the reinforcement learning method with curiosity mechanism performs well in such tasks and achieves the most advanced results at present.(3)A knowledge map question answering system with curiosity mechanism and reinforcement learning is designed and implemented,which can answer related questions raised by users online and verify the usability and performance of the system through a series of tests.To sum up,this paper proposes a knowledge map question answering method based on reinforcement learning,and proposes to use curiosity mechanism to solve the sparse reward problem in reinforcement learning,and designs and implements a knowledge map question answering system to meet the practical needs.