Research On Multi-hop Question-answering Model In Agriculture Based On Knowledge Graph Embedding And Path Reasoning

In the era of big data,it is very important to help users find the information they need from massive data quickly and accurately.In recent years,question-answering technology has been highly concerned and deeply studied by experts,and intelligent question-answering system has been widely used in many fields.In the field of agriculture,question-answering system can help farmers or agricultural researchers to obtain agricultural knowledge conveniently and quickly,which can be used to guide agricultural production and research.Researchers at home and abroad have proposed some question-answering methods based on agricultural knowledge graph.At present,most of these methods can only be used to answer simple questions.However,in practical application scenarios,users tend to ask complex multi-hop questions,and simple knowledge questions and answers usually can not meet the needs of users,so the multi-hop question-answering model with strong reasoning ability has become one of the urgent problems to be solved in this field.In view of the above problems,this thesis takes agricultural domain knowledge as the background,and studies the multihop question-answering method based on knowledge graph.The main research work of this thesis is as follows:(1)Design a Compl Ex＿CLA of knowledge graph embedding representation model which integrates comparative learning and attention.In the training stage of Compl Ex embedding model,contrast loss and attention are added to shorten the semantic distance between related entities and entity-relation couples in different triples,so as to better capture the high-order relations between entities,thereby improving the representation ability of knowledge graph embedding.The link prediction ability of the proposed model is verified by experiments on multiple data sets of different scales,and the embedded representation learning training of three knowledge graph data sets is completed at the same time,which lays a foundation for the research of downstream task question-answering model.(2)Design a multi-hop question-answering model KEPRQA based on knowledge graph embedding and path reasoning.In the knowledge graph embedding module,a Compl Ex＿CLA model is used to fully learn the semantic information representation of the knowledge graph,and in the answer reasoning stage,knowledge graph embedding and weak supervision training of path reasoning are used to realize multi-hop question-answering task reasoning by predicting the implicit intermediate relations in the knowledge graph.Experiments on private agricultural data sets and several public data sets have achieved good results.(3)A question-answering prototype system for rice diseases and insect pests based on knowledge graph was designed and implemented.Based on the knowledge graph multi-hop question-answering model proposed in this thesis,the basic function of rice diseases and insect pests question-answering system is designed and implemented.On the basis of completing the conventional question-answering,the system realizes the function of semiartificial dynamic update and maintenance of system data,so that system managers can manage and maintain the knowledge graph data and question sentence data in time,and the system test tests the practicability and reliability of the system.