C3N

The code is related to the paper below: Improving Multimodal Fake News Detection by Leveraging Cross-modal Content Correlation.

Data

The two datasets used for this project are publicly accessible, and their links are provided below.

weibo

twitter

In this project, due to upload size limitations for some data, the files in /data/weibo/processed/crops/ used can be downloaded via Google Drive.

Requirements

We train our model on Python 3.7.0 and Pytorch 1.8.0. And your environment should have some packages as follows:

clip==1.0
cn_clip==1.5.1
importlib_metadata==6.0.0
langdetect==1.0.9
matplotlib==2.2.4
numpy==1.21.6
opencv_python==4.7.0.68
pandas==1.1.5
Pillow==9.5.0
scikit_learn==1.0.2
seaborn==0.12.2
torchtext==0.15.2
tqdm==4.64.1
transformers==4.25.1

Run

After installing the environment, in the code directory, modify the save paths in main.py, process_image_weibo.py, and process_text_weibo.py.

First, execute process_text_weibo.py to obtain the complete text preprocessing results.

Next, run process_image_weibo.py to obtain the image preprocessing results.

Run main.py to perform the training process.