This repository is the official implementation of the COLING 2025 paper: SubRegWeigh: Effective and Efficient AnnotationWeighing with Subword Regularization
Put following data on ./data folder
- CoNLL 2003 original: https://www.clips.uantwerpen.be/conll2003/ner/
- CoNLL++(CoNLL CW): https://github.com/ZihanWangKi/CrossWeigh
- CoNLL++(CoNLL 2020): https://github.com/ShuhengL/acl2023_conllpp
- SSL2: https://huggingface.co/datasets/stanfordnlp/sst2
./SubRegWeigh/scripts/run.sh
Note: For detailed experimental results, please refer to our paper.
| Models | Time | CoNLL CW | CoNLL 2020 |
|---|---|---|---|
| RoBERTA LARGE | - | 95.27 | 94.80 |
| +SubRegWeigh (K-means) | 5:21 | 95.45 | 94.96 |
| LUKE LARGE | - | 95.92 | 95.29 |
| +SubRegWeigh (K-means) | 6:36 | 96.12 | 95.31 |
Note: This citation refers to the preprint version on arXiv. The official COLING 2025 proceedings citation will be updated once it becomes available.
@article{tsuji2024subregweigh,
title={SubRegWeigh: Effective and Efficient Annotation Weighing with Subword Regularization},
author={Tsuji, Kohei and Hiraoka, Tatsuya and Cheng, Yuchang and Iwakura, Tomoya},
journal={arXiv preprint arXiv:2409.06216},
year={2024}
}