“Fake News” in Your Feed

twitter bird logo

How reliable is your Twitter feed? It might not be as honest as you think. A recent article in the MIT Technology Review highlighted a recent competition hosted by DARPA (the Defense Advanced Research Projects Agency) which had teams attempting to identify which tweets were generated by people, and which were generated by “bots”, programs created to automatically post tweets. According to the article, Twitter admitted that some 8 percent of its user accounts were “Twitterbots”, which amounts to around 23 million user accounts. During the course of the competition, teams analyzed some 4 million tweets from thousands of accounts over four weeks.

The winning team, from the social media analytics company Sentimetrix, was able to devise an algorithm that correctly identified 39 out of 40 bots in the competition dataset. The team was also successfully able to sort the bots into various categories, including “influence bots”. Influence bots are programmed to sway Twitter-based discussions in a particular direction, for instance, in favor of support of a controversial politician or viewpoint.  The winning team outlined their strategies in a paper published in the IEEE Computer Magazine.

Another project, hosted by Indiana University, is called the Observatory on Social Media (OSoMe). The OSoMe lab has developed a number of interesting tools to analyze information on social media. They developed one called “Bot or Not”, which will apply a score to a Twitter user account, depending on how likely it is to be a bot account.

Not all bots are bad – there’s even a Twitterbot, @BotDetectorBot, programmed to analyze other Twitter accounts and highlight ones with many fake followers. A paid service called Twitter Audit lets individuals run audits of their Twitter followers, and has an option to block bot accounts.

While Twitter bots can be relatively harmless, a large number are malicious or illegal. Some spam users, while others direct users to malicious websites. While there has been improvement in detecting and blocking automated tweets, it remains a significant challenge to maintain a useful and reliable social media ecosystem.