Rule-based text data augmentation is widely used for NLP tasks due to its simplicity. However, this method can potentially damage the original meaning of the text, ultimately hurting the performance of the model. To overcome this limitation, we propose a straightforward technique for applying soft labels to augmented data. We conducted experiments across seven different classification tasks and empirically demonstrated the effectiveness of our proposed approach. We have publicly opened our source code for reproducibility.
@article{arxiv.2402.05591,
title = {SoftEDA: Rethinking Rule-Based Data Augmentation with Soft Labels},
author = {Juhwan Choi and Kyohoon Jin and Junho Lee and Sangmin Song and Youngbin Kim},
journal= {arXiv preprint arXiv:2402.05591},
year = {2024}
}