Characterization and detection of inauthentic coordinated behaviours in strategic information operations on Twitter
In this work there is an analysis based on historical inauthentic Information Operations (IOs) detected by Twitter and removed from the social, used as a ground truth, to study patterns of inauthentic and harmful coordination and to compare them with a dataset of Twitter messages recovered using the Twitter APIs and based on the most used hashtags in the selected IOs. These can be considered authentic and harmless, because surely they don't come from users involved in the campaign, banned by Twitter. From the archive, the selected IOs come from Honduras and United Arab Emirates (UAE), which used different information strategies. The first was focused only on political aspects and is based on retweets; the second concerned both political and religious contents and exploited classic conversations. First of all highly coordinated communities (HCCs) are extracted from the whole datasets, composed by the union of authentic and inauthentic tweets, to analyze indexes and behaviours and to see if communities of banned users are isolated from the others. Then, some classifiers are applied to each single user of the datasets. The final purpose is to try to predict through supervised machine learning if users are involved in an inauthentic information operation or not, with high precision score.
