A Word is Worth A Thousand Dollars: Adversarial Attack on Tweets Fools Stock Predictions
This addresses a security risk for investors and financial models reliant on social media data, representing an incremental advance in adversarial attack research.
The paper tackles the vulnerability of stock prediction models to adversarial attacks on social media text, demonstrating that their method can achieve consistent success rates and cause significant monetary loss in trading simulations by perturbing tweets.
More and more investors and machine learning models rely on social media (e.g., Twitter and Reddit) to gather real-time information and sentiment to predict stock price movements. Although text-based models are known to be vulnerable to adversarial attacks, whether stock prediction models have similar vulnerability is underexplored. In this paper, we experiment with a variety of adversarial attack configurations to fool three stock prediction victim models. We address the task of adversarial generation by solving combinatorial optimization problems with semantics and budget constraints. Our results show that the proposed attack method can achieve consistent success rates and cause significant monetary loss in trading simulation by simply concatenating a perturbed but semantically similar tweet.