AT-BERT: Adversarial Training BERT for Acronym Identification Winning Solution for SDU@AAAI-21
Abstract
Acronym identification focuses on finding the acronyms and the phrases that have been abbreviated, which is crucial for scientific document understanding tasks. However, the limited size of manually annotated datasets hinders further improvement for the problem. Recent breakthroughs of language models pre-trained on large corpora clearly show that unsupervised pre-training can vastly improve the performance of downstream tasks. In this paper, we present an Adversarial Training BERT method named AT-BERT, our winning solution to acronym identification task for Scientific Document Understanding (SDU) Challenge of AAAI 2021. Specifically, the pre-trained BERT is adopted to capture better semantic representation. Then we incorporate the FGM adversarial training strategy into the fine-tuning of BERT, which makes the model more robust and generalized. Furthermore, an ensemble mechanism is devised to involve the representations learned from multiple BERT variants. Assembling all these components together, the experimental results on the SciAI dataset show that our proposed approach outperforms all other competitive state-of-the-art methods.
Cite
@article{arxiv.2101.03700,
title = {AT-BERT: Adversarial Training BERT for Acronym Identification Winning Solution for SDU@AAAI-21},
author = {Danqing Zhu and Wangli Lin and Yang Zhang and Qiwei Zhong and Guanxiong Zeng and Weilin Wu and Jiayu Tang},
journal= {arXiv preprint arXiv:2101.03700},
year = {2021}
}
Comments
Accepted to SDU @ AAAI 2021, 8 pages, 3 figures