A tweet can wreak havoc in a few hundred characters, as demonstrated in April 2013 when someone hacked the Associated Press Twitter account and claimed that explosions at the White House had injured President Barack Obama. There were no explosions — and Obama was fine — but the Dow Jones average sank by 100 points in two minutes. Stock markets swiftly recovered once the truth came out. Twitter, however, remained a breeding pool for false information.
Some of Twitter’s rumors are true. The discovery of the Higgs boson leaked through Twitter before its official announcement in 2012. Others, of course, are false and far more pernicious, such as conspiracy theories about the recent high school shooting in Parkland, Fla.
A new study in Science quantifies the spread of Twitter rumors. Previous research tracked rumors after specific events, like the false information that swirled around the Boston Marathon bombing. The study published Thursday is more wide-ranging: A team of researchers at MIT tracked falsehoods and truths using a database of every tweet written from 2006 to 2017.
False news moved through Twitter “farther, faster, deeper and more broadly” than the truth, said Sinan Aral, a professor of information technology at MIT who studies social media networks.
The study authors aimed to be apolitical in distinguishing what was true or false. They deliberately did not use the phrase fake news, and they outsourced claims of veracity to Snopes.com, Politifact.com and four other independent fact-checking organizations. “The central concept of this paper is veracity,” Aral said. “I say that boldly because I know it’s difficult to make a claim like that.”
On any given news item, the fact-checking organizations were in agreement between 95 and 98 percent of the time. “We don’t know of any other way to get a more rigorous data set,” than using independent fact-checkers that are nearly unanimous, Aral said.
Aral and his colleagues fished through the Twitter database for a specific type of reply tweet. If a Twitter user had linked to a fact-checking organization’s Web page, either to support or debunk a spreading rumor, the game was afoot. From that link, the researchers backtracked through the retweet chain, which they called a cascade, to find the rumor’s origin. This method of cataloguing tweets produced 126,000 true or false cascades involving 3 million people.
By almost all metrics, false cascades outpaced true ones. Even the farthest-reaching true rumors rarely spread to more than 1,000 people. But the top 1 percent of falsehoods routinely had audiences of 1,000 to 100,000 people, the study authors reported. Politics got the most attention among true and false rumors, they discovered, representing 45,000 of the 126,000 cascades. False political rumors had a particularly high peak during the 2016 election.
A single account was responsible for starting 4,700 false rumors. Aral declined to identify it, citing the conditions that Twitter imposed when sharing the data set.
But he offered generalities about who propagates false news. A false rumor cascade was more likely to begin with a young, unverified account with a small number of followers. False news spread farther “despite these characteristics, not because of them,” Aral said.
In a statistical model that kept variables like Twitter followers and account age equal, falsehoods were 70 percent more likely to be retweeted than the truth. The study authors hypothesized that falsehoods contain more novelty than truth. To that end, they measured the “information uniqueness” of rumors and discovered that false rumors were more likely to contain new, but wrong, information.
“It’s easier to be novel when you’re unconstrained by reality,” Aral said. Similarly, the researchers identified common themes in the phrasing of replies to false rumors — users more frequently expressed words associated with disgust and surprise when they commented on untruths.
“The authors are very honest with the interpretation of their results: They cannot claim any causality between novelty and endorsement, but they provide convincing evidence that novelty plays an important role in spreading fake information,” said Manlio De Domenico, a scientist at the Bruno Kessler Foundation’s Center for Information and Communication Technology in Italy who tracked how the Higgs boson rumor spread on Twitter.
The study authors said automated Twitter accounts, or bots, were not to blame for the faster spread of false rumors. Using techniques to identify bots, they determined that software-run accounts spread falsehoods and truths equally. Put another way, only human activity could explain the preferential spread of false news.
De Domenico said his recent research, studying bot activity during the Catalan independence referendum, also supports the idea that software accelerates “true and false news at the same rate.” He said he suspects that “bots are designed just to increase anarchy in online social systems.”
There are no easy fixes to solve false news on the Internet, Aral said. The act of labeling news as false, some early studies have shown, might increase its spread. And that idea raises its own problems: Who, for instance, would be responsible for stamping tweets with the label of truth? Not Twitter, apparently, according to one company executive who recently testified before British officials that “we are not the arbiters of truth.”
“We need more research,” Aral said. “There hasn’t been any real large-scale study of this phenomenon up until now.”