EMNLP 2021: A Novel Global Feature-Oriented Relational Triple Extraction Model based on Table Filling.
The main requirements are:
- python 3.6
- torch 1.7.0
- tqdm
- transformers 3.5.1
- bert4keras
-
Get pre-trained BERT model Download BERT-BASE-CASED and put it under
./pretrained
. -
Train and select the model
python run.py --dataset=WebNLG --train=train --rounds=4
python run.py --dataset=WebNLG_star --train=train --rounds=2
python run.py --dataset=NYT24 --train=train --rounds=3
python run.py --dataset=NYT24_star --train=train --rounds=2
python run.py --dataset=NYT29 --train=train --rounds=3
- Evaluate on the test set
python run.py --dataset=WebNLG --train=test --rounds=4
python run.py --dataset=WebNLG_star --train=test --rounds=2
python run.py --dataset=NYT24 --train=test --rounds=3
python run.py --dataset=NYT24_star --train=test --rounds=2
python run.py --dataset=NYT29 --train=test --rounds=3
Parts of our codes come from bert4keras.