This is the source code of IJCNN 2023 paper TieFake: Title-Text Similarity and Emotion-Aware Fake News Detection (TieFake).

Start

To run the code in this repo, you need to have Python>=3.9.6, PyTorch>=1.9.0 Other dependencies can be installed using the following commands:

pip install -r requirements.txt download datasets clean datasets and save them into folder Data,such as:

--Data

--politifact_images

--xx.jpg

......

--gossipcop_images

--xx.jpg

......

--politifact_train.tsv

--politifact_test.tsv

--gossipcop_train.tsv

--gossipcop_test.tsv

run bert_training.py to train bert in our datasets run resnest101_training.py to train resnest_101 in our datasets run main.py to train fusion_model

Datasets

Complete dataset cannot be distributed because of Twitter privacy policies and news publisher copy rights. The dataset includes fake&real from dataset FakeNewsNet,including Politifact and Gossipcop.

After we clean the datasets, the statistics of the dataset is shown below:

| News Articles | #Fake News| #True News | #Total News |

| Politifact | 161 | 205 | 366 |

| Gossipcop | 4927 | 11693 | 21620 |

If you use the code in your project, please cite the following paper: IJCNN'23 (PDF)

@inproceedings{guo2023TieFake,
  title={TieFake: Title-Text Similarity and Emotion-Aware Fake News Detection},
  author={Guo, Quanjiang and Kang, Zhao and Tian, Ling and Chen, Zhouguo},
  booktitle={Proceedings of the IEEE International Joint Conference on Neural Networks 2023},
  year={2023}
}

Please email to guochance1999@163.com for other inquiries.