get's Introduction

GET

Source code and datasets for the paper "Evidence-aware Fake News Detection with Graph Neural Networks".

Requirements

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

pip install -r requirements.txt

Data

We utilize two widely used datasets.

Usage

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

sh run_politifact.sh (on the PolitiFact dataset)

Citation

Please cite our paper if you use the code:

@misc{xu2022evidenceaware,
      title={Evidence-aware Fake News Detection with Graph Neural Networks},
      author={Weizhi Xu and Junfei Wu and Qiang Liu and Shu Wu and Liang Wang},
      year={2022},
      eprint={2201.06885},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Acknowledgement

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.

Recommend Projects