softEDA is a variant of EDA (Easy Data Augmentation) (EMNLP 2019). softEDA introduces soft label on augmented data by applying label smoothing. The figure below shows the structure of softEDA compared to traditional EDA. For more information, please refer to the paper!
pip install -r requirements.txt
# Check utils/arguments.py and modify path arguments
python total_search.py
If you found this work helpful for your future research, please consider citing this work:
@inproceedings{choi2023softeda,
title={SoftEDA: Rethinking Rule-Based Data Augmentation with Soft Labels},
author={Choi, Juhwan and Jin, Kyohoon and Lee, Junho and Song, Sangmin and Kim, YoungBin},
booktitle={ICLR 2023 Tiny Papers},
year={2023}
}