ZJUKLAB at SemEval-2021 Task 4: Negative Augmentation with Language Model for Reading Comprehension of Abstract Meaning

ZJUKLAB at SemEval-2021 Task 4: Negative Augmentation with Language Model for Reading Comprehension of Abstract Meaning

This repository is the official implementation of the ACL 2021 (Semeval Workshop) paper ZJUKLAB at SemEval-2021 Task 4: Negative Augmentation with Language Model for Reading Comprehension of Abstract Meaning which is the 4th rank system of the SemEval 2021 Task4.

Contributors

Student: Xin Xie, Nanxiang Chen, Xiang Chen and so on

Supervisor: Ningyu Zhang, Huajun Chen.

Requirements

To install requirements:

• python >= 3.7

• torch >= 1.6

• transformers==3.3.1

pip install -r requirements.txt

preprocess

To get the NAL answers, we use the models without fine-tuning to get the enhanced dataset.

python ./dataset/preprocess.py

Training

post-training

Firstly, pretrain the ALBERT,RoBERTa models to fit the in-domain text.

./scripts/pretrain_model.sh

fine-tuning

Secondly, fine-tuning the model with the followed scripts.

./scripts/run_deberta.sh
./scripts/run_albert.sh
./scripts/run_roberta.sh

ensemble

Finally, we get the best model files and ensemble them with weighted voting (weighted by the acc at dev set).

./scripts/get_answer/save_answer.sh

Evaluation

run the command below to get the result files.

./scripts/get_answer/save_answer.sh

Results

Our model achieves the following performance on :

Papers for the Project & How to Cite

If you use or extend our work, please cite the following paper:

@inproceedings{DBLP:conf/semeval/XieCCWZDC21,
  author    = {Xin Xie and
               Xiangnan Chen and
               Xiang Chen and
               Yong Wang and
               Ningyu Zhang and
               Shumin Deng and
               Huajun Chen},
  editor    = {Alexis Palmer and
               Nathan Schneider and
               Natalie Schluter and
               Guy Emerson and
               Aur{\'{e}}lie Herbelot and
               Xiaodan Zhu},
  title     = {{ZJUKLAB} at SemEval-2021 Task 4: Negative Augmentation with Language
               Model for Reading Comprehension of Abstract Meaning},
  booktitle = {Proceedings of the 15th International Workshop on Semantic Evaluation,
               SemEval@ACL/IJCNLP 2021, Virtual Event / Bangkok, Thailand, August
               5-6, 2021},
  pages     = {810--819},
  publisher = {Association for Computational Linguistics},
  year      = {2021},
  url       = {https://doi.org/10.18653/v1/2021.semeval-1.108},
  doi       = {10.18653/v1/2021.semeval-1.108},
  timestamp = {Mon, 22 Nov 2021 09:21:39 +0100},
  biburl    = {https://dblp.org/rec/conf/semeval/XieCCWZDC21.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}