Bot or Not: Could a New Platform Help Schools Detect Fake Twitter Profiles?

By Leo Versel — December 04, 2017 3 min read
  • Save to favorites
  • Print

As fake and suspicious profiles continue to circulate on social media platforms, posing challenges for educators and students, two undergraduate students have created a program designed to flag illegitimate accounts., a Google Chrome extension available via the browser’s web store, enables individuals to reveal and log information about suspicious Twitter users.

Ash Bhat, co-founder of Robhat Labs, developed Botcheck with Rohan Phadte, his friend and roommate at the University of California, Berkeley, who he has known since middle school.

After downloading from the Google Chrome web store, users will see a button next to any Twitter account, which they can click to determine whether that account shows signs of being a political propaganda bot.

In an online post on Medium, Bhat and Phadte define these bots as “semi-automated or automated Twitter accounts” that typically identify themselves as real people and endorse politically-polarizing content by re-tweeting false information.

“For us, this problem was something we had to run after and try to solve,” Bhat said in an interview. “If technology can create a dichotomy in how content is created, it should also be able to provide a solution for that problem.”

The two students set out to analyze where fake news online was originating and discovered social media sites, particularly Twitter, were among the main sources, Bhat explained. In researching and classifying Twitter users, they found a group of accounts on the social media platform outstripped others in the frequency of tweets spreading false information. Many of those profiles were automated bots, Bhat said.

The issue of suspicious Twitter accounts just one of the factors complicating K-12 teachers’ efforts to educate their students in media literacy. Teachers are increasingly expected to help students become more skilled in identifying credible and dubious information, navigating internet hoaxes, and evaluating the credibility of partisan advocacy and posts disguised as objective news, as EdWeek writer Benjamin Herold explained in a December 2016 story.

Bhat said Botcheck can teach K-12 students and teachers to not take information they see online as fact, simply because it is published on a website. He also emphasized the importance of considering primary sources on content-aggregating sites such as Wikipedia in teaching students media literacy.

“The internet is incredibly valuable for information and has created a vast quantity, but this has come at the cost of quality,” Bhat said. “Now that anyone can publish a secondary source, primary sources have become even more valuable.”

Before creating Botcheck, Bhat said he and Phadte tried to determine the political biases of each Twitter profile. Some accounts, he said, did not fit into either a left or right-leaning group. Both creators of the extension then generated a bot training data set of about 200 characteristics that frequently appeared in false profiles.

Thousands of Suspicious Profiles

After developing an algorithm based on this data set, Bhat and Phadte developed a machine learning model, or a form of artificial intelligence that can analyze trends in data, and in this case, identify the veracity of a Twitter profile. The pair eventually built an application integrating the model.

“The extension makes calls to a server, which hosts our machine learning model and makes classifications of the accounts based on our data set,” Bhat explained. As the platform collects more information, Bhat and Phadte are actively updating that model.

Bhat and Phadte previously tackled the phenomenon of “fake news” in May 2017, launching another artificial intelligence bot together for Robhat Labs, “NewsBot AI,” which determined the validity and biases of news articles shared online. has been available to users since October 30. Botcheck’s page on the Chrome Web Store shows that 3,451 users have downloaded the extension, as of December 1. According to Bhat, users have found over 50,000 classifications of suspicious profiles in that time, which have all been logged to the extension’s server.

Bhat said that he and Phadte have not received a response from Twitter about their platform. But he said the response from Botcheck’s users has been “overwhelmingly positive.”

“Over the past four weeks, we have been re-training new machine learning models every other day based on user feedback,” Bhat said. That work is paying off: the system’s accuracy in detecting bots, he said, has gone from 93 percent to nearly 96 percent.

See also:

A version of this news article first appeared in the Digital Education blog.