Skip to content

Latest commit

 

History

History
72 lines (56 loc) · 2.32 KB

README.md

File metadata and controls

72 lines (56 loc) · 2.32 KB

GET

model

This is the code for the www'22 Paper: Evidence-aware Fake News Detection with Graph Neural Networks.

Usage

We utilize two widely used datasets.

You can run the commands below to train and test our model on Snopes Dataset.

python MasterFC/master_get.py --dataset="Snopes" \
                             --cuda=1 \
                             --fixed_length_left=30 \
                             --fixed_length_right=100 \
                             --log="logs/get" \
                             --loss_type="cross_entropy" \
                             --batch_size=32 \
                             --num_folds=5 \
                             --use_claim_source=0 \
                             --use_article_source=1 \
                             --path="formatted_data/declare/" \
                             --hidden_size=300 \
                             --epochs=100 \
                             --num_att_heads_for_words=5 \
                             --num_att_heads_for_evds=2 \
                             --gnn_window_size=3 \
                             --lr=0.0001 \
                             --gnn_dropout=0.2 \
                             --seed=123756 \
                             --gsl_rate=0.6

You can also simply run the bash script.

sh run_snopes.sh

or

sh run_politifact.sh (on the PolitiFact dataset)

Requirements

We use Pytorch 1.9.1 and python 3.6. Other requirements are in requirements.txt.

pip install -r requirements.txt

Citation

Please cite our paper if you use the code:

@inproceedings{xu2022evidence,
  title={Evidence-aware fake news detection with graph neural networks},
  author={Xu, Weizhi and Wu, Junfei and Liu, Qiang and Wu, Shu and Wang, Liang},
  booktitle={Proceedings of the ACM web conference 2022},
  pages={2501--2510},
  year={2022}
}

Acknowledge

The general structure of our codes inherites from the open-source codes of MAC, we thank them for their great contribution to the research community of fake news detection.