| 注册
首页|期刊导航|计算机技术与发展|基于BERT的猕猴桃病虫害命名实体识别

基于BERT的猕猴桃病虫害命名实体识别

文青青 李金星 王虎 李莉婕 赵泽英 郭雷风

计算机技术与发展2025,Vol.35Issue(5):214-220,7.
计算机技术与发展2025,Vol.35Issue(5):214-220,7.DOI:10.20165/j.cnki.ISSN1673-629X.2024.0404

基于BERT的猕猴桃病虫害命名实体识别

Named Entity Recognition of Kiwifruit Pests and Diseases Based on BERT

文青青 1李金星 1王虎 2李莉婕 2赵泽英 2郭雷风1

作者信息

  • 1. 新疆农业大学计算机与信息工程学院,新疆乌鲁木齐 830052||中国农业科学院农业信息研究所,北京 100080
  • 2. 贵州省农业科技信息研究所,贵州 贵阳 550006
  • 折叠

摘要

Abstract

Aiming at the difficulties in the field of kiwifruit pests and diseases,such as the scarcity of training data,the complex and variable entity structure,the variety of entity types,and the uneven distribution of entities,a BERT model based on data enhancement was introduced.Firstly,the kiwifruit pest and disease dataset KIVINER,which contains 11 entity categories,was constructed after analyzing the information of six collected books,and then in order to effectively expand the kiwifruit pest and disease dataset,two data enhancement methods,namely,sentence restructuring and language model-based generation,were used.In the preliminary study,the performance of five models,namely IDCNN-CRF,BERT-BiLSTM-CRF,RoBERT-BiLSTM-CRF,ALBERT,and BERT,were tested on the original kiwifruit pest and disease dataset.The experimental results showed that the BERT model achieved F1 values of over 90%on all 10 classes of entities,with an overall F1 value of 93.4%.Subsequently,the effects of different data enhancement strategies on the performance of the BERT model in the kiwifruit pest and disease named entity recognition task were explored.The experimental results showed that the F1 values of the sentence restructuring approach and the language model-based approach were improved to 94.29%and 94.11%,respectively.Combining the two data enhancement methods resulted in the highest precision,recall and F1 values of 94.12%,95.17%and 94.64%,respectively,while the model also achieved good results on the public dataset Resume,indicating that the model has some generalization ability.

关键词

猕猴桃病虫害/数据增强/BERT/句子重组/语言模型生成/命名实体识别

Key words

kiwifruit pests and diseases/data enhancement/BERT/sentence restructuring/language model generation/named entity recog-nition

分类

信息技术与安全科学

引用本文复制引用

文青青,李金星,王虎,李莉婕,赵泽英,郭雷风..基于BERT的猕猴桃病虫害命名实体识别[J].计算机技术与发展,2025,35(5):214-220,7.

基金项目

贵州省科技支撑项目(黔科合支撑[2023]一般189) (黔科合支撑[2023]一般189)

贵州省科研机构创新能力建设专项资金(黔科合服企[2021]15号) (黔科合服企[2021]15号)

计算机技术与发展

1673-629X

访问量5
|
下载量0
段落导航相关论文