融合数据增强的互花米草入侵关联要素实体识别方法  

Entity recognition method of spartina alterniflora invasion associated factors fused data augmentation

作　　者：李忠伟[1] 张文丰 李永[1] 李明轩[1] LI Zhong-wei;ZHANG Wen-feng;LI Yong;LI Ming-xuan(College of Oceanography and Space Informatics,China University of Petroleum(East China),Qingdao 266400,China;Qingdao Institute of Software,College of Computer Science and Technology,China University of Petroleum(East China),Qingdao 266400,China)

机构地区：[1]中国石油大学(华东)海洋空间与信息学院,山东青岛266400 [2]中国石油大学(华东)青岛软件学院计算机科学与技术学院,山东青岛266400

出　　处：《计算机工程与设计》2025年第2期603-609,共7页Computer Engineering and Design

基　　金：自主创新科研计划基金项目(理工科)-战略专项(22CX01004A)。

摘　　要：为解决互花米草入侵领域的训练数据匮乏,存在实体特征提取不准确的问题,提出一种融合数据增强的互花米草入侵关联要素识别深度学习模型。将训练数据采用同类实体随机交叉互换的方法进行数据增强,利用BERT预训练获得互花米草入侵关联要素的上下文信息;使用BiLSTM进一步提取特征,利用CRF得到实体的标签约束。通过对比不同模型在自建数据集上的精确率、召回率和F1分数,验证了该模型在互花米草入侵领域实体识别的有效性。To solve the insufficient training data and inaccurate entity feature extraction problems in spartina alterniflora invasion field,a deep learning model of spartina alterniflora invasion associated factors recognition fused data augmentation was proposed.The training data were processed by DA using the method of random cross interchanged similar entities,and the context information of Spartina alterniflora invasion associated factors was obtained through pre-training model.The BiLSTM was used to further extract the features,and the tag constraints were obtained through CRF.By comparing the accuracy,recall rate and F1 score of different models on the self-built data set,the effectiveness of the model in spartina alterniflora invasion field was verified.

关 键 词：命名实体识别 互花米草入侵 深度学习 数据增强 预训练模型 双向长短期记忆网络 条件随机场 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]