基于FastBert的水稻病虫害实体关系抽取研究  

Relationship Extraction of Entities About Rice Diseases and Insect Pests Based on FastBert

作　　者：周烨 徐向英[1,2] 章永龙 陈佳云[1] 汪洪江 Zhou Ye;Xu Xiangying;Zhang Yonglong;Chen Jiayun;Wang Hongjiang(College of Information Engineering,Yangzhou University,Yangzhou 225012 China;Joint International Research Laboratory of Agriculture and Agri-Product Safety,the Ministry of Education of China,Yangzhou 225127 China)

机构地区：[1]扬州大学信息工程学院,江苏扬州225012 [2]扬州大学教育部农业与农产品安全国际合作联合实验室,江苏扬州225127

出　　处：《南京师范大学学报（工程技术版）》2023年第1期33-38,共6页Journal of Nanjing Normal University(Engineering and Technology Edition)

基　　金：教育部农业与农产品安全国际合作联合实验室开放课题项目(JILAR-KF202007);扬州大学交叉学科基金项目(yzuxk202008);扬州市市校合作专项项目(YZ2021150)。

摘　　要：针对水稻病虫害知识图谱构建所需实体和关系,提出了一种基于FastBert模型的中文实体关系抽取方法.首先,在中文语料收集的基础上,使用Hanlp工具和农业词典提取了与水稻病虫害相关的领域实体,并依据实体间关系的特点定义了病虫害别名、为害部位、为害地区、防治方法等7种类型.然后,在词嵌入和句子嵌入的基础上通过FastBert模型实现水稻病虫害关系的抽取.该模型与Robert、Electra、Distilbert等其它Bert相关模型的关系抽取结果比较显示,基于FastBert模型的中文水稻病虫害关系抽取效果更好,模型获得的实体间关系F1值达0.72,模型精度达0.69.该方法为中文农业病虫害知识图谱的自动化构建提供了参考.A FastBert model based Chinese entity relationship extraction method is proposed to extract the entities and relationships required for rice pest and disease knowledge graph.First of all,on the basis of Chinese corpus collected,a tool named Hanlp and a agricultural dictionary are used to extract the domain entities related to rice diseases and insect pests.According to the characteristics of the relationship between entities,seven types of diseases and pests are defined,such as alias,harm parts,suffer region,prevention and treatment,etc.Based on word embedding and sentence embedding,the extraction of the relation of rice diseases and insect pests is realized through the FastBert model.And the results are compared with those of other Bert related models.It shows that the FastBert model is better than other Bert related models in the relationship extraction task of entities in the Chinese corpus of rice diseases and insect pest.The F1 value obtained by the FastBert model is 0.72,and the accuracy of the model is 0.69.This method provides a reference for automated construction of Chinese knowledge map of agricultural pests and diseases.

关 键 词：水稻病虫害 知识图谱 关系抽取 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]