Detecting inorganic financial campaigns on Twitter

Online financial content is widespread on social media, especially on Twitter. The possibility to access open, real-time data about stock market information and firms' public reputation can bring competitive advantages to industry insiders. However, as many studies extensively demonstrated before, manipulative campaigns by social bots do not spare the financial sector either. In this work, we show that the more viral a stock is on Twitter, the more that virality is artificially caused by social bots. This result is also confirmed when considering accounts suspended by Twitter instead of bots. Starting from this finding, we then propose two methods for detecting the presence and the extent of financial disinformation on Twitter, via classification and regression. Our systems exploit hundreds of features to encode the characteristics of viral discussions, including features about: participating users, textual content of shared posts, temporal patterns of diffusion, and financial information about stocks. We experiment with different combinations of algorithms and features, achieving excellent results for the detection of financial disinformation (F1 = 0.97) and promising results for the challenging task of estimating the extent of inorganic activity within financial discussions (R2 = 0.81, MAE = 4.9%). Our compelling results pave the way for the deployment of novel systems for protecting against financial disinformation.