Skip to content

[Paper] Code for the EMNLP2023 (Findings) paper "Global Structure Knowledge-Guided Relation Extraction Method for Visually-Rich Document"

License

Notifications You must be signed in to change notification settings

chenxn2020/GOSE

Repository files navigation

🦢GOSE

Awesome License: MIT

👋 News!

Quick Links

Setup

We check the reproducibility under this environment.

  • Python 3.8.18
  • CUDA 11.1

To run the codes, you need to install the requirements:

git clone https://github.com/chenxn2020/GOSE.git
cd GOSE

conda create -n gose python=3.8
conda activate gose
pip install -r requirements.txt

Model Preparation

We utilize LayoutXLM and LiLT as our backbone. You can download the models and place them under the GOSE/.

Train GOSE

We provide example scripts for explaining the usage of our code. You can kindly run the following commands.

  • Language-specific Fine-tuning

# Current path:  */GOSE
bash standard.sh
  • Multilingual fine-tuning

# Current path:  */GOSE
bash multi.sh

Acknowledgment

The repository benefits greatly from unilm/layoutlmft and LiLT. Thanks a lot for their excellent work.

Citation

If our paper helps your research, please cite it in your publication(s):

@article{DBLP:journals/corr/abs-2305-13850,
  author       = {Xiangnan Chen and
                  Juncheng Li and
                  Duo Dong and
                  Qian Xiao and
                  Jun Lin and
                  Xiaozhong Liu and
                  Siliang Tang},
  title        = {Global Structure Knowledge-Guided Relation Extraction Method for Visually-Rich
                  Document},
  journal      = {CoRR},
  volume       = {abs/2305.13850},
  year         = {2023},
  url          = {https://doi.org/10.48550/arXiv.2305.13850},
  doi          = {10.48550/ARXIV.2305.13850},
  eprinttype    = {arXiv},
  eprint       = {2305.13850},
  timestamp    = {Mon, 05 Jun 2023 15:42:15 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-2305-13850.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}

About

[Paper] Code for the EMNLP2023 (Findings) paper "Global Structure Knowledge-Guided Relation Extraction Method for Visually-Rich Document"

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published