Real-Time Classification of Market-Moving Events on Social Media: Distinguishing Natural Occurrences from Coordinated Misinformation Campaigns

Social media are platforms emerging as valuable sources of breaking news. Information gathering and evaluation speed has become increasingly critical in financial markets. Recent events around geo-political tensions have shown that local eyewitnesses often share first-hand accounts of significant developments on platforms like X up to 30 minutes before mainstream media coverage begins. This creates an advantage for market participants, like market-makers, who can effectively extract social media signals.

However, the rise of coordinated misinformation campaigns, particularly those by nation-states, presents a significant challenge to the approach mentioned above to extracting advantageous information. These campaigns, if detected as a (false-positive) signal, can break a trading strategy. This project aims to introduce a method to distinguish between true and synthetical clusters of posts.

This research addresses this challenge by developing a framework for real-time classification of social media event signals. We focus on creating a probability estimation system that updates with each new post addition, trying to classify the event mentioned in the post as a natural (true) event or synthetical (false, misinformation campaign). The most significant part of the project focused on creating a dataset of verified geo-political events and confirmed misinformation campaigns as the behavior of misinformation campaigns constantly evolves; therefore, no publicly available quality dataset exists for our approach.

The dataset is still in development and will eventually be published by synoptic. What we initially found is that while the sentence structure and variance significantly differs between the two classes, an interesting and unexpected finding is that misinformation campaigns seem to be harder to trace back to initial posts from local eyewhitnesses in the event location.