Why Twitter gossip is such a headache

Twitter is a valuable source of fascinating commentary, replete with links to important journalism and otter videos. Twitter is also a flaming chamber of unchecked gossip.

At least, that’s according to a newly published study in PLOS One. A team of researchers, led by computer scientist Arkaitz Zubiaga of the University of Warwick, wanted to track the life cycle of a social media rumor.

To do this, they found a new way to analyze a randomly selected subset of more than 300 rumor threads (nearly 5,000 tweets) associated with nine recent news stories. Their objective? To see how Twitter users either spread, supported, or denied rumors that are later proven true or false.

The rumors the researchers studied had to do with a smattering of issues, from the unrest in Ferguson to the Charlie Hebdo shooting to whether Prince would actually play a secret show in Toronto. (He didn’t, though people lined up anyway.)

What the researchers learned was that “rumors that are ultimately proven true tend to be resolved faster than those that turn out to be false.”

More specifically, true rumors were corroborated within two hours of being introduced on Twitter, while false rumors tended to take about 14 hours to get debunked.

The researchers chalked up the time gap to the legwork it takes to do a thorough debunking. “False stories take much longer on average to be debunked because counter-evidence takes longer to come to light,” said lead author Zubiaga. “It takes longer to investigate that something has not occurred than it does to prove something has indeed occurred.” It’s also possible that false rumors have other characteristics that makes them harder to debunk.

Another interesting finding from this study is that even esteemed news organizations tend to release early, unverified reports, using hedging language like “reportedly,” or attributing the report to other sources such as the police, Zubiaga said. “After all, this facilitates releasing later corrections without being accountable for the error.”

But people tend to share whatever news organizations report, helping rumors that look like trusted news to spread quicker.

The researchers proposed at least one solution in the paper: developing machine learning tools that can flag reports as rumors, so users are aware before spreading the dubious information even further. (There are already similar tools out there, like TwitterTrails, which measures how widely a social media rumor has spread and how skeptical users should be about it.)

For now, Zubiaga said, don’t believe everything you read on Twitter. “The main implication of false rumors taking longer to be debunked than true rumors to be confirmed is that the longer it takes a rumor to be resolved, the more cautious you should be, as it is more likely to be false.”