Design And Realization Of Question Answering System For Crop Pests And Diseases Based On Knowledge Graph

The average annual area of pests and diseases occur in China is 460-508 million hm2,which has a huge impact on the production of various crops.The use of scientific knowledge and technology of control methods has an important impact on the yield and quality of crops.At present,the dissemination of knowledge about crop diseases and pests has problems such as inefficiency,distortion,and backward means.Therefore,it is of great significance to build an accurate,efficient and fast Q&A system for pest and disease Control.In this study,based on the data of disease and pest,combined with knowledge ontology,a knowledge graph of crop diseases and insect pests was constructed.In this study,transforming the task of understanding the knowledge problem of pests and diseases into the task of problem template classification.Finally integrate the Q&A model to realize the pests and diseases question answering system.The results are as follows:1.Construction of the knowledge graph of crop diseases and insect pests.According to the characteristics of professional knowledge of crop diseases and insect pests,a top-down construction model is adopted,combined with the knowledge structure of crop diseases and insect pests,a 5-layer ontology is designed,and the field and scope of disease and insect pest knowledge are determined.According to the knowledge characteristics of pests and diseases,construct a knowledge framework of entities,relationships and attributes,and realize 12 types and 1065 kinds of knowledge graph of pests and diseases,including 6594 kinds of pests and diseases entities of crops.There are 7044 relations and 6994 triples;on this basis,11 types of question templates are designed.Constructed 155,916 natural language questions with 132 categories.2.Building a question answering model based on natural language processing.Based on the knowledge graph of crop pests and diseases,Han LP Chinese natural language processing tool was used to identify the named entities of crop pests and diseases,and 1114 dictionaries of crop pests and diseases were formed;Using natural language processing technology,NBC based,BERT based and Text CNN based models are constructed and and tested.The results show that the accuracy of the NBC,BERT and BERT+Text CNN models respectively reached 96.8%,93.0%,and 99.6%.Among them,the most stable model was NBC.The evaluation metrics were all above 95%.The optimal model was BERT+Text CNN.The precision rate,recall rate and F1 value were all 99.5%.The above shows that it is feasible to realize problem classification based on machine learning and to use knowledge graph to realize knowledge matching of crop diseases and insect pests.3.The realization of crop disease and pest Q&A system.Based on the knowledge graph,combined with the crop pests and diseases Q&A model,using Web service and Flask framework technology,a 3-layer architecture of Q&A system is designed,and the system is constructed,which realizes crop knowledge graph and question and answer model.The integration with the question and answer service,combined with the named entity recognition technology,realizes the accurate recommendation of the knowledge of crop diseases and insect pests.