基于对比学习的多标签层级情感文本分类方法  

Multi-label Hierarchical Sentiment Text Classification Method Based on Contrastive Learning

作　　者：朱珍元[1] 祁鹏 ZHU Zhenyuan;QI Peng(Department of Information Management,Anhui Vocational College of Police Officers,Hefei Anhui 230031,China;iFLYTEK Company Limited,Hefei Anhui 230088,China)

机构地区：[1]安徽警官职业学院信息管理系,安徽合肥230031 [2]科大讯飞股份有限公司,安徽合肥230088

出　　处：《鞍山师范学院学报》2024年第6期75-82,共8页Journal of Anshan Normal University

基　　金：安徽省高等学校自然科学研究重点项目(2022AH052939);安徽警官职业学院教学研究重点项目(2022YJJYXM11).

摘　　要：传统的情感分类方法在处理多标签情感分类任务时,往往难以有效捕捉标签之间的依赖关系,导致分类性能不理想.针对多标签情感分类任务中标签之间复杂关系难以处理的问题,提出一种基于对比学习的多标签层级情感文本分类方法.通过引入GoEmotions情绪标签体系,并基于图神经网络进行层次标签表征,结合对比学习策略,优化模型的表示空间,从而提升分类精度.通过在GoEmotions和ISEAR数据集上进行实验,验证了所提出方法的有效性.实验结果表明,该方法在多标签情感分类任务中的表现显著优于现有基线方法.When dealing with multi-label emotion classification tasks,traditional emotion classification methods are often difficult to capture the dependency between labels effectively,which leads to poor classification performance.In order to solve the difficult problem of complex relationships between labels in multi-label emotion classification task,a multi-label hierarchical emotion text classification method based on contrast learning is proposed.By introducing GoEmotions labeling system,hierarchical labeling representation based on graph neural network,combined with comparative learning strategy,the aim is to optimize the representation space of the model and improve the classification accuracy.Experiments were conducted on GoEmotions and ISEAR datasets to verify the effectiveness of the proposed method.The experimental results show that the proposed method performs significantly better than the existing baseline method in multi-label emotion classification tasks.

关 键 词：多标签情感分类 对比学习 图神经网络 层次标签表征 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]