EPiDA: An Easy Plug-in Data Augmentation Framework for High Performance Text Classification

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4742 - 4752 July 10-15, 2022 ©2022 Association for Computational Linguistics EPiDA: An Easy Plug-in Data Augmentation Framework for High Performance Text Classification Minyi Zhao1 Lu Zhang1 Yi Xu1 Jiandong Ding2 Jihong Guan3 Shuigeng Zhou1∗ 1Shanghai Key Lab of Intelligent Information Processing, and School of Computer Science, Fudan University, China 2Alibaba Group 3Tongji University, China 1,2{zhaomy20, l_zhang19, yxu17, jdding, sgzhou}@fudan.edu.cn 3jhguan@tongji.edu.cn Abstract Recent works have empirically shown the ef- fectiveness of data augmentation (DA) for NLP tasks, especially for those suffering from data scarcity. Intuitively, given the size of generated data, their diversity and quality are crucial to the performance of targeted tasks. However, to the best of our knowledge, most existing meth- ods consider only either the diversity or the quality of augmented data, thus cannot fully tap the potential of DA for NLP. In this paper, we present an easy and plug-in data augmentation framework EPiDA to support effective text clas- sification. EPiDA employs two mechanisms: relative entropy maximization (REM) and con- ditional entropy minimization (CEM) to control data generation, where REM is designed to en- hance the diversity of augmented data while CEM is exploited to ensure their semantic con- sistency. EPiDA can support efficient and con- tinuous data generation for effective classifier training. Extensive experiments show that EP- iDA outperforms existing SOTA methods in most cases, though not using any agent network or pre-trained generation network, and it works well with various DA algorithms and classifi- cation models. Code is available at https: //github.com/zhaominyiz/EPiDA. 1 Introduction Data augmentation (DA) is widely-used in clas- sification tasks (Shorten and Khoshgoftaar, 2019; Feng et al., 2021; Zhang et al., 2021). In computer vision (CV), (Krizhevsky et al., 2012; Chatfield et al., 2014; Szegedy et al., 2015) adopt strate- gies like flipping, cropping, tilting to perform DA. In natural language processing (NLP), (Xie et al., 2017; Coulombe, 2018; Niu and Bansal, 2018; Wei and Zou, 2019) find that native augmentation skills such as spelling errors, synonym replacement, deleting and swapping, can bring considerable per- formance improvement. All these methods use ∗Corresponding author. various transformations for data augmentation, but they do not achieve equal success in different NLP tasks (Yang et al., 2020). Sometimes, they fail to guarantee semantic consistency, and may even bring semantic errors that are harmful to classifica- tion. The reason lies in that data augmentation for NLP is in discrete space, so it can easily incur large deviation of semantics (e.g. in sentiment classifica- tion task, deleting emotional words from a sentence will make its meaning completely different). Generally, given the size of generated data, their diversity and quality are crucial to the perfor- mance of targeted tasks (Ash et al., 2019). Re- cent works have begun to emphasize the diversity or quality of augmented data. For example, in CV, AA (Cubuk et al., 2019), Fast-AA (Lim et al., 2019) and LTA (Luo et al., 2020) employ …

Content truncated to 500 words