Twitter Bots Multiclass Classification Using Bot-Like Behavior Features Open Access Deposited
Downloadable ContentDownload PDF
Bots on Twitter are accounts that are controlled by computer programs, automatically producing content, and interacting with other accounts. These programs are turned on and off without following a pattern, making them hard to identify. Using previous work that identifies bot accounts bot-like behavior features, we identified features that are more relevant to some bot types than others. In this thesis, we propose a novel bot type classification method by using bot-like behavior features. Multiclass classification is our proposed idea for this project. We use the output data from the bot-like behavior to train a MaxEnt classifier to identify 6 different classes (5 bots, 1 human). We collect our test dataset to match the structure of our training set except for labels, then use our classifier to test it. Moreover, we analyze an additional holdout set to test the polarity of bot classes in vaccine topics on Twitter.