README.md

DeMis: Data-efficient Misinformation Detection using RL

Resources for misinformation detection on Twitter. This repo is the official resource of the following paper.

• DeMis: Data-efficient Misinformation Detection using Reinforcement Learning, ECML-PKDD 2022.
• [Link][PDF][Slide]

📚 Data Sets

The data sets about COVID-19 misinformation on Twitter presented in our paper are available below.

• COMYTH (weather & home-remedies) - [Datasets]
• COVIDLies - [Paper]

🚀 Pre-trained Models

We release our models for misinformation detection on Twitter trained using DeMis method. There are three models trained on three COVID-19 misinformation data sets separately. All trained misinformation detection models are available on my Google Drive 🤗 so you can download models via PyTorch and use it for prediction right away!!!

• DeMis-COMYTH-W (trained on COVID-weather data)
• DeMis-COMYTH-H (trained on COVID-home-remedies data)
• DeMis-COVIDLies (trained on COVIDLies data)

⚙️ Usage

You can use the code to do any of the following purposes: (1) preprocessing tweets, (2) running the model for prediction, or (3) training the model. We tested in pytorch v1.10.1 and transformers v4.18.0.

Usage 1: Preprocessing tweets

Specify the input and output filepaths in the shell script run_tweet_preprocessing.sh and run the following command.

sh scripts/run_tweet_preprocessing.sh

Usage 2: Run the detection model for classification

Specify the model path, input and output filepaths in the shell script run_detector.sh. Note that you can download the models from the section above and try running it with the following command, or you can train a new detection model using DeMis in the next section.

sh scripts/run_detector.sh

Usage 3: Train detector using DeMis

Run a sample script to train using COVID-weather data. See README to learn more on how to train with your data.

sh scripts/run_sample_entire_demis_pipeline.sh

✏️ Citation

If you feel our paper and resources are useful, please consider citing our work! 🙏

@inproceedings{kawintiranon2022demis,
  title     = {DeMis: Data-efficient Misinformation Detection using Reinforcement Learning},
  author    = {Kawintiranon, Kornraphop and Singh, Lisa},
  booktitle = {Proceedings of the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML-PKDD)},
  year      = {2022},
  publisher = {Springer}
}

🛠 Throubleshoots

Create an issue here if you have any issues loading models or data sets.