Dual-CAN

Code for EACL 2023 Findings Entity-Aware Dual Co-Attention Network for Fake News Detection

How to Run

1. Put data in the same directory (bigru, BERT_based, Dual_CAN_model) of the code.
2. For Bi-GRU and Dual-CAN model, need to preprocess the data by running

python preprocess.py --data_dir data

3. train and test the model

python main.py --data_dir data --output_dir output
python main.py --data_dir data --output_dir output --mode 1

How to Create data

• Follow FakeNewsNet (https://github.com/KaiDMML/FakeNewsNet) and CoAID (https://github.com/cuilimeng/CoAID) instructions to download the data.
• The id we use for the experiments are in data directory.
  • For GossipCop, the data_id is the name of the directory. We only use tweet without retweet.
  • For CoAID, we use the first version (05-01-2020). The dataid is the (label+id). Again, We only use tweet without retweet.

Preprocessing Data of GossipCop

• Execute the code in data/GossipCop

python preprocess.py
python preprocess_tweets.py

Preprocessing Data of CoAID

• Execute the code in data/CoAID

python get_data.py
python preprocess.py

Dependencies

python 3.7
torch 1.7.1
tensorflow 2.4.1
pytorch-lightning 1.2.3
spacy 3.0.5
seqeval 1.2.2
tqdm
numpy
pandas
scikit_learn
tagme
wikipediaapi