Tweets Need A Credibility Rating And This Lie Detector Algorithm Could Help

Next Story

LogMeIn’s Dropbox Competitor Cubby Reveals Pricing; Stays Competitive At $7 Per Month For 100 GB

wikipedian_protester (2)

The Newtown school shooting tragedy turned social media channels into another disastrous rumor mill. Ryan Lanza and his friends suffered a flood of hate messages after vigilante facebookers attacked him following false reports by CNN (the shooter was Adam Lanza, Ryan’s brother). Actor Morgan Freeman had to deny a viral violence-in-media rant widely attributed to him (although, this may not be as bad as when rumors of his death were circulated). To combat vicious falsehoods, social media updates need a credibility rating. A new lie detecting algorithm that correctly identified truthful tweets 86% of the time could mark a giant leap forward toward that goal.

Following up on a 2010 Yahoo! study, the updated paper to be published in next month’s Internet Research finds that users themselves are the best sources of credibility. Specifically, it builds upon known characteristics [PDF] of credible tweets:

  1. longer updates
  2. include URLs
  3. be tweeted by users with more follower counts
  4. tweets have a more negative tone
  5. contain swear words
  6. 6. contain more frowny emoticons :-(

Interestingly enough, users themselves tend to question potential rumors, making their aggregate skepticism a key indicator of credibility. Authors Carlos Castillo, Marcelo Mendoza, and Barbara Poblete explain,

These results show that the propagation of tweets that correspond to rumors differs from tweets that spread news because rumors tend to be questioned more than news by the Twitter community. Notice that this fact suggests that the Twitter community works like a collaborative filter of information. This result suggests also a very promising research line: it could posible [sic] to detect rumors by using aggregate analysis on tweets.

In the updated study, as reported by Slate, when presented with a random false or truthful tweet, the algorithm correctly identified the truthful ones 86% of the time.

As these predictors become more reliable over time, web apps, such as Tweetdeck, or even Twitter itself, could rank tweets by their credibility, hopefully sparing the public from more vicious rumors. Long ago, Wikipedia recognized this with its famous “[citation needed]” tag. For all our sakes, its about time Twitter and Facebook developed their own version.

[Image Credit: XKCD]