/CFFN

Primary LanguagePython

CFFN

Code for the MMAsia 2023 paper [Cross-modal Consistency Learning with Fine-grained Fusion Network for Multimodal Fake News Detection]

Dataset

The datasets used in our paper are Twitter and Weibo. The code of dataset preprocessing is listed in resource\dataset folder:

  • Weibo folder for Weibo dataset
  • Mediaeval folder for Twitter dataset

pretrained model

For Weibo dataset, bert-base-chinese is needed and moving it to the folder resource/bert.

For Twitter dataset, bert_base is needed and moving it to the folder resource/bert.

Dependencies

  python 3.8.1
  pytorch-pretrained-bert==0.6.2
  timm==0.4.12
  numpy==1.23.5
  tensorboard==2.12.0
  torch 1.9.0 + cu11.1
  torchvision==0.10.0
  pandas==1.5.3
  scikit-learn==0.24.1
  gensim==4.3.1
  jieba==0.42.1
  tqdm==4.64.1
  python-json-logger==2.0.7
  transformers==3.3.1

Running the Code

For Weibo dataset：

cd ./weibo
bash train.sh

For Twitter dataset：

cd ./Mediaeval
bash train.sh