xinyadu / memory_docie

Dynamic Global Memory for Document-level Argument Extraction

Code and data for paper the ACL 22 paper (link).

Dependencies

• pytorch=1.6
• transformers=3.1.0
• pytorch-lightning=1.0.6
• spacy=3.0 # conflicts with transformers
• pytorch-struct=0.4
• sentence-transformers=2.1.0

Datasets

• WikiEvents (included in this repo)

Running

• Normal data setting

  Train:./scripts/train_kairos.sh Test: ./scripts/test_kairos.sh

• Adversarial data setting

  Train:./scripts/train_kairos_adv.sh Test: ./scripts/test_kairos_adv.sh

Cite

If you use our code or data/outputs, please cite:

@InProceedings{memory_ie,
  author = {Du, Xinya and Li, Sha and Ji, Heng},
  title = {Dynamic Global Memory for Document-level Argument Extraction},
  booktitle = {Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics},
  year = {2022},
}