We consider a tweet concatenation attack that 'retweet' semantically similar tweets to fool financial forecasting models. This repository is the official implementation of our paper.

Run the following command to install requirements:

pip install -r requirements.txt

The preprocessed dataset, used resources and trained victim models used in the paper are publicly available at: dataset. resource and trained models.

Alternatively, run the following command to download and decompress the resources. It also creates the necessary folders used in the training and attacking.

bash downloader.sh

To run the attack and look into the results, run the following command with model arguements han, stocknet, tweetgru or tweetlstm:

bash attack.sh han

It conducts concatenation attack with perturbation of replacement for various budget via joint optimization. The results are saved in /log/attack. The attack uses our trained models. Change the arguments in the script to implement different attacks.

To train the 4 victim models from the scratch, run this command with a model argument:

bash train.sh han

It trains the models with the same hyperparameters used in the paper. The training logs and checkpoints are save in /log/train and checkpoints respectively.

• Effect of attack budget on Attack Success Rate (ASR)

• Impact of the attack on portfolio PnL: trading simulation with initial value $10000 shows our attack causes additional loss of $3200 (32%) over two years.

@article{xie2022advtweet, 
  title={A Word is Worth A Thousand Dollars: Adversarial Attack on Tweets Fools Stock Prediction},
  author={Xie, Yong and Wang, Dakuo and Chen, Pin-Yu and Jinjun, Xiong and Liu, Sijia and Koyejo, Oluwasanmi},
  journal={Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},
  year={2022}
}