Bot-hunting Twitter bot sniffs out bogus political tweets

If you’ve ever wondered if the hashtag heavy, politically far afield Twitter user you’re about to eviscerate in 140 characters is totally delusional or merely not quite sentient, well, wonder no more.

On Wenesday, Quartz just launched a handy Twitter bot known as @probabot_ to sort so much automated chaff from Twitter’s proverbial fields of wheat. Probabot identifies political tweets and then analyzes them using something called the Botometer, a tool developed by the Indiana University Network Science Institute and the Center for Complex Networks and Systems Research.

This account has a Botometer score of 75.0%, which suggests it is very likely to be a bot (or bot-assisted) https://t.co/ft4OZBRH4U

— probabot (@probabot_) October 23, 2017

The latter project lets users analyze Twitter accounts with machine learning to see the probability of if an account is automated. Accounts can also fall into the category of “bot-assisted” which suggests a combination of bot-style autopilot and human intervention.

Using the Botometer API, Probabot scans accounts on potential axes of botliness, including when they tweet, if the content of their tweets is positive or negative in sentiment, who they tag in their tweets and how often, and who else is in their network. Probabot creator Keith Collins explained to TechCrunch how his bot is programmed to avoid false positives through things like avoiding verified users and organization accounts, which sometimes express some bot-like behavior. As Collins explained, accounts that are rated above 48% by the Botometer are flagged as potential bots, while anything over 60% rates as a “likely” bot.

Twitter doesn’t provide a lot of insight around just how widespread its bot problem might be, stating only that about 5% of accounts are “fake or spam accounts.” Even if that number is an accurate estimate, given how prolific some of these bots are they could command an outsized influence on the political discourse on the platform. As Quartz notes, one of the likely bots on Probabot’s list has tweeted over one million times since April 2016, something that even the most loyal Twitter user couldn’t pull off.

This account has a Botometer score of 74.0%, which suggests it is very likely to be a bot (or bot-assisted) https://t.co/eH7pVpDIhJ

— probabot (@probabot_) October 25, 2017

As one of the researchers behind the Botometer, Emilio Ferrara, Ph.D., explained to Quartz, “We found that approximately 15% of the users active in the political conversation one month prior to the 2016 election were likely bots… and they were responsible to about one in five such political tweets (nearly 20%).”

“The usage of bots has migrated during the last year or so from political propaganda to inciting chaos and divisiveness in this country,” Ferrara told TechCrunch. “The spread of misinformation by means of bots is currently mostly concerned with divisive messaging around social issues, and the impact on our society is more consequential now than ever.”

The Probabot is a useful project, but it’s worth remembering that it can only determine if something is probably a bot. So far Probabot has produced a running list of 35 suspected accounts, so it could be worth checking back to see what they have in common as the data set expands. Ultimately the tools tie into the Botometer creators’ greater research aims around the study of bots on social platforms, much of which they’ve published. In 2017, it’s stuff well worth reading.