ahren09/finerfact

PyTorch implementation for the FinerFact model in the AAAI 2022 paper Towards Fine-Grained Reasoning for Fake News Detection

FinerFact

This is the PyTorch implementation for the FinerFact model in the AAAI 2022 paper Towards Fine-Grained Reasoning for Fake News Detection (Arxiv).

@article{jin2021towards,
  title={Towards Fine-Grained Reasoning for Fake News Detection},
  author={Jin, Yiqiao and Wang, Xiting and Yang, Ruichao and Sun, Yizhou and Wang, Wei and Liao, Hao and Xie, Xing},
  journal={arXiv preprint arXiv:2110.15064},
  year={2021}
}

The implementation is based on HuggingFace Transformers and KernelGAT.

Installation

• Run conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch. conda is preferred over pip due to its stability on Windows

Instruction to run code

• Take politifact as an example. Make sure you have put the following training and test files under data/.
  • Train_bert-base-cased_politifact_130_5.pt
  • Test_bert-base-cased_politifact_130_5.pt
• If the Train_*.pt and Test_*.pt files are not present, you can run preprocess/preprocess.py to split the training data (e.g. bert-base-cased_politifact_130_5.pt) into Train_*.pt and Test_*.pt. You can download the data here
• Download the files for pretrained BERT model and put them under bert_base/. You should have the following 3 files in bert_base/:
  • pytorch_model.bin
  • vocab.txt
  • bert_config.json
• make sure you have set the root path given by get_root_dir() in utils/utils to your own data path of fake_news_data/. Mine is root = "C:\\Workspace\\FakeNews\\fake_news_data" on Windows and root = "../../fake_news_data"
• run the train.py file using kgat/ as the working directory:
  • python train.py --outdir . --config_file P.ini, or
  • python train.py --outdir . --config_file G.ini

finerfact's Issues

How to test the trained model and how to prepare my dataset to run this model?

Thank you!
The script for runing test.py and the missing file.
The format of the datasets in "C:\Workspace\FakeNews\fake_news_data".

Request for Access to Dataset for Reproducing Results in "FinerFact" Paper

Dear Dr. Jin and Dr. Liao,

I hope this email finds you well. My name is Zhang Weijian, and I am a Ph.D. candidate in Computer Science at the University of Macau, currently conducting research on fake news detection.

I recently came across your paper titled "Towards Fine-Grained Reasoning for Fake News Detection" on arXiv (https://arxiv.org/abs/2110.15064) and found it highly relevant and insightful for my research. I am particularly interested in reproducing the results presented in your paper to further understand and build upon your work.

I have accessed the code provided in your GitHub repository (https://github.com/Ahren09/FinerFact). However, I noticed that the dataset required to run the experiments is not included. Given the recent restrictions imposed by Twitter on API access, I am unable to independently download the necessary Twitter data to recreate the dataset.

Therefore, I am writing to kindly request if you could share the dataset used in your study. Access to this dataset would be invaluable for my research, allowing me to accurately replicate your findings and contribute to the ongoing research in this field. I assure you that any data shared will be used strictly for academic purposes and in accordance with data privacy and ethical guidelines.

Thank you very much for your time and consideration. I look forward to your positive response.

Best regards,

Zhang Weijian
Ph.D. Candidate, Computer Science
University of Macau
[email protected]

Could you please tell me how to prepare the dataset?Is that do not include the claim-evidence graph construct part?

lack utils.utils_case_study file

for kgat/test.py
from utils.utils_case_study import load_case_study is error.
in utils file, there is no utils.utils_case_study.py file