This code implements SAFE: Similarity-Aware Multi-modal Fake News Detection model.
We use FakeNewsNet dataset and provide our data in this link. For the latest verision of FakeNewsNet, please directly check out: https://github.com/KaiDMML/FakeNewsNet.
We use Show and Tell to abstract the content of images.
We embed words use pre-trained word vectors glove.840B.300d and the embedding tool SIF. The computation of glove.840B.300d word map is time-consuming, in order to provide more convenience we upload (words, We), which is the result of data_io.getWordmap(wordfile) in SIF. Please check embedding branch for the modified code and embedding results.
- Python 3.7
- TensorFlow 2.2
- xlwt
- nltk
pip install -r requirements.txt
python3 helper.py
python3 train.py
python3 test.py
If you use this code for your research, please cite our paper:
@inproceedings{zhou2020multimodal,
title={SAFE: Similarity-Aware Multi-modal Fake News Detection},
author={Zhou, Xinyi and Wu, Jindi and Zafarani, Reza},
booktitle={Pacific-Asia Conference on Knowledge Discovery and Data Mining},
pages={354--367},
year={2020},
organization={Springer}
}
If you have any question, please contact zhouxinyi@data.syr.edu.