The original project is tested under the following environments:

python==3.8.2
torch==1.8.1+cu111
pytorch-lightning==1.5.0
transformers==4.10.0
scikit-learn==1.2.1

Please download the .jsonl files and store them under ./data.

The datasets can be downloaded using the links below (requires format conversion, code can be found in ./scripts/:

• Essays (Stab and Gurevych, 2017): link
• AbstRCT (Mayer et al., 2020): link
• ECHR (Poudyal et al., 2020): link
• CDCP (Park and Cardie, 2018; Niculae et al., 2017): link
• AMPERE++ link

Or you can directly download the processed data from link.

To train a standard supervised relation extraction model on CDCP:

SEED=1
DOMAIN="cdcp"

python -m argument_relation_transformer.train \
  --datadir=./data \
  --seed=${SEED} \
  --dataset=${DOMAIN} \
  --ckptdir=./checkpoints \
  --exp-name=demo-${DOMAIN}_seed=${SEED} \
  --warmup-steps=5000 \
  --learning-rate=1e-5 \
  --scheduler=constant \
  --max-epochs=15 \
  --window-size=10 \
  --cl=1e-3