细粒度烟草病虫害知识图谱构建研究

打开文本图片集

关键词：烟草病虫害防治；知识图谱；信息抽取；命名实体识别；关系抽取中图分类号：S435.72 文献标识码：A 文章编号：2095-5553（2025）12-0040-10

Abstract：Inresponse to thecurrent problems of complexity of pest anddisease types in the tobacco pest domainand the lack of specialized platforms forasisting preventionandcontrol，a fine-grained knowledge graph of tobaccopestsand diseases is constructed by using both top-down and botom-up approaches.In the tobacc disease and pest knowledge repository，six typesof entities and five types of relationshipsare defined.Basedonthis，a tobaco pestand disease knowledge graph dataset，TobaccoKG，isconstructed through entity-relation annotation.An ERNIE3.O—BiGRU— GCN—MHSA—CRF model，which incorporates entity position information，isemployed forNamed EntityRecognition （NER）tasks in tobacco pest and disease domain，while an ERNIE3.O—Atention—TextCNN model with integrated maskedprompts isused forRelationExtraction（RE）tasks.Theexperimentalresults indicate thattheproposedmodels achieve F1 scores of 95.54% and 96.51% for NER and RE tasks，respectively，greatly ensures the quality and reliability of the knowledge graph.In the Neo4j graph database，the knowledge graph of tobaco pests and diseases is stored and visualized toachieve acurate queryof pestand disease preventionandcontrol informationandinteligent asistance for tobacco farmers to effectively diagnose and treat pests and diseases.

Keywords：tobacco disease and pestcontrol；knowledgegraph；information extraction；named entity recognition; relation extraction

0 引言

响烟草的产量和品质。（剩余15614字）