MDFEND: Multi-domain Fake News Detection

This is an official implementation for MDFEND: Multi-domain Fake News Detection which has been accepted by CIKM2021.

Dataset

The splited dataset (i.e., train, val, test) are in the MDFEND-Weibo21/data folder.

You can have access to the original dataset of Weibo21 only after an "Application to Use Weibo21 for Fake News Detection" has been submitted.

Code

Requirements

Refer to requirements.txt

You can run pip install -r requirements.txt to deploy the environment quickly.

pretrained_model

You can download pretrained model (Roberta) from https://drive.google.com/drive/folders/1y2k22iMG1i1f302NLf-bj7UEe9zwTwLR?usp=sharing and move all the files in the folder into the path MDFEND-Weibo21/pretrained_model/chinese_roberta_wwm_base_ext_pytorch.

Data Preparation

After you download the Weibo21 dataset (the way to access is described here), move the train.pkl, val.pkl and test.pkl into the path MDFEND-Weibo21/data.

Run

You can run this model through:

python main.py --model_name mdfend --batchsize 32 --lr 0.0007

Reference

Nan Q, Cao J, Zhu Y, et al. MDFEND: Multi-domain Fake News Detection[C]//Proceedings of the 30th ACM International Conference on Information & Knowledge Management. 2021: 3343-3347.

or in bibtex style:

@inproceedings{nan2021mdfend,
  title={MDFEND: Multi-domain Fake News Detection},
  author={Nan, Qiong and Cao, Juan and Zhu, Yongchun and Wang, Yanyan and Li, Jintao},
  booktitle={Proceedings of the 30th ACM International Conference on Information \& Knowledge Management},
  pages={3343--3347},
  year={2021}
}