I Spot a Bot: Building a binary classifier to detect bots on Twitter

It has been estimated that up to 50% of the activity on Twitter comes from bots [1]: algorithmically-automated accounts created to advertize products, distribute spam, or sway public opinion. It is perhaps this last intent that is most alarming; studies have found that up to 20% of the Twitter activity related to the 2016 U.S. presidential election came from suspected bot accounts, and there has been evidence of bots used to spread false rumors about French presidential candidate Emmanuel Macron and to escalate a recent conflict in Qatar [1]. Detecting bots is necessary in order to identify bad actors in the “Twitterverse” and protect genuine users from misinformation and malicious intents. This has been an area of research for several years, but current algorithms still lag in performance relative to humans [2]. Given a Twitter user’s profile and tweet history, our project was to build a binary classifier that identifies a given user as “bot” or “human.” The end-user application for a classifier such as this one would be a web plug-in for the browser that can score a given account in real-time (See page 5 for mock-ups). All of the raw inputs required to classify a public Twitter account via our algorithm are available for download from the Twitter API; in fact, our check_screenname.py program is a working prototype that uses the API to classify a given Twitter user handle within seconds. It is our opinion that a product like this is sorely needed for the average Twitter consumer.