SoftEDA: Rethinking Rule-Based Data Augmentation with Soft Labels

Juhwan Choi, Kyohoon Jin, Junho Lee, Sangmin Song, Youngbin Kim
2/8/2024
cs.CLcs.AI

No official code found. Best implementation: IIPL-CAU/SoftEDA (5 stars, last commit 2023-07-25).

Abstract

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.

View on arXivView PDF

Code Implementations(1)

IIPL-CAU/SoftEDACommunity100%
51PythonFeb 14, 20233 years ago

Similar Papers

AutoAugment Is What You Need: Enhancing Rule-based Augmentation Methods in Low-resource Regimes

86% match1 repo32024

STA: Self-controlled Text Augmentation for Improving Text Classifications

80% match1 repo52023

Don't Waste a Single Annotation: Improving Single-Label Classifiers Through Soft Labels

80% match2023

Adverb Is the Key: Simple Text Data Augmentation with Adverb Deletion

78% match1 repo2024

Enhancing Effectiveness and Robustness in a Low-Resource Regime via Decision-Boundary-aware Data Augmentation

78% match2024

Batch Aggregation: An Approach to Enhance Text Classification with Correlated Augmented Data

77% match2025

Cite this paper

@article{choi2024softeda,
  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},
  year   = {2024},
  eprint = {2402.05591},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2402.05591v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2402.05591.svg)](https://codeofpaper.com/paper/2402.05591)

Discussion