Design and Optimization of Loss Function in Fine-grained Sentiment and Emotion Analysis
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
万方数据
在细粒度情感分析和情绪分析数据集中,标签之间的相关性和标签分布的不均匀性非常突出.类别标签分布不均匀,标签之间存在相关性容易影响学习模型的性能.针对这一问题,该文受计算机视觉领域中的 Circle loss 方法的启发,将梯度衰减、成对优化、添加余量引入损失函数来优化深度学习模型的性能.该方法可以很好地与预训练模型相结合,不需要修改骨干网络.与当前最新的经典方法相比,该方法在 SemEval18 数据集上Jaccard系数、micro-F1、macro-F1 分别提升了 1.9%、2%、1.9%;在GoeEmotions数据集上Jaccard系数、micro-F1、macro-F1 分别提升了 2.6%、1.9%、3.6%.实验表明,该文提出的损失函数对情感分析和情绪分析问题具有显著的提升作用.
In fine-grained sentiment and emotion analysis tasks,the label correlation and imbalanced label distribu-tion are popular among samples.Inspired by circle loss in computer version,we develop a loss function model to handle these issues by employing gradient decay,pair optimization and margin.This loss function model is easily a-dapted to suit pre-trained networks without modifying the backbone structures.Compared with the current state-of-the-art results,our loss function model could improve Jaccard similarity coefficient,micro-F1,and macro-F1 val-ues by 1.9%,2%,and 1.9%,respectively,in SemEval18 dataset;and by 2.6%,1.9%,and 3.6%,respectively,in GoEmotions dataset.
sentiment analysisemotion analysispair optimizationloss function
NETL
NSTL
万方数据
