UniArk

The repository for the NAACL 2024 paper "UniArk: Improving Generalisation and Consistency for Factual Knowledge Extraction through Debiasing".

Thanks for your attention. The code and dataset ParaTrex has been available.

Environments

To set up the required enviroments, use:

conda env create -f environment.yml 

Usages

To run the experiments:

bash train_bert.sh

Data

The packed LAMA dataset is available at [Zenodo]. s If you use our packed up data, please download it and unzip it in the data/ folder in the root directory.

Acknowledge

We thank the implementation of P-tuning, which inspires some code in this repo.

Citation

@article{yang2024uniark,
  title={UniArk: Improving Generalisation and Consistency for Factual Knowledge Extraction through Debiasing},
  author={Yang, Yijun and He, Jie and Chen, Pinzhen and Guti{\'e}rrez-Basulto, V{\'\i}ctor and Pan, Jeff Z},
  journal={arXiv preprint arXiv:2404.01253},
  year={2024}
}