Lies Have Longevity on the Internet

Despite correction after correction, internet hoaxes and misinformation can live a long and oftentimes damaging life on social media, according to joint research from the University of Washington and Northwest University that was presented at the 2014 iConference in Berlin.

In an analysis of three false rumors spread over Twitter during coverage of the 2013 Boston Marathon bombing, the research team found that lies spread quickly—but corrections did not. For instance, a rumor that an eight-year-old runner was killed was retweeted more than 90,000 times over the five days following the blast, despite the fact that it could be easily debunked (marathon rules prohibit runners under 18). Twitter’s crowdsourced fact-checking networks failed to correct that rumor with the same viral fervor.

“Even after a correction comes out, the false rumor persists, and at an orders-of-magnitude higher level,” said Robert M. Mason, a co-author on the study and a founding member of the University of Washington’s Social Media Lab. “That was a surprise. We expected something different.”

twitter-screengrab_2048x1152
Lies on Twitter tend to be shorter, end in question marks or exclamation points, and be propagated by Twitter users with relatively low follower counts.

Mason’s study contradicts previous research indicating that social media users are just as good at debunking lies as they are at creating them. A study by Yahoo Research labs that tracked a small number of rumors that cropped up on Twitter in the wake of the 2010 Chilean earthquake found much higher correction rates than Mason’s study and concluded that “false rumors tend to be questioned [on social media] much more than confirmed truths.”

It’s unclear as to exactly why the studies conflict, but it’s worth noting that both studies only investigated a small number of rumors—ten in total between the two studies—meaning that results may potentially be more conclusive upon examination of a bigger number of lies. Thankfully (or not), social media has more than enough fodder for researchers who are up for the data crunching.

In the future, users may not have to manually suss fact from fiction at all. In 2012, for the same Chilean earthquake tweet, scientists created an automated classifier that detected social media lies with 86% efficacy. After analyzing 16 features of tweets, researchers identified certain trends. Lies, for instance, are more likely to be shorter (though the paper published in the journal, Internet Research, doesn’t mention by how much), end in question marks or exclamation points, and be propagated by Twitter users with relatively low follower counts. Using that data, the classifier was able to pinpoint tweets statistically most likely to be false, though it’s unclear how the method would work with more nuanced or partially true statements.

We don’t quite have a reliable way to mathematically squash e-gossip before it blossoms, but we do have one valuable weapon in the war against internet rumors—the correction. Like lies, corrections can go viral too, said Jeff Hemsley, co-author of the book “Going Viral” and an assistant professor at Syracuse University’s School of Information Studies. Even if they don’t in all cases, they still serve a valuable function in preventing the rumor mill from continuing to churn. “At some point the power of the rumor is overwhelmed by the people who say, ‘This is boloney,’ and the rumor dies down,” Hemsley said. “The fact kind of kills the story because often, [the truth] is not very interesting.”

Until science offers a way to stop internet rumors before they start, we should focus on perfecting the human engineering, rather than the algorithmic engineering, that drives social media, said Dr. Mason. “Remind the people who are using the system that retweets can be harmful.”