Trending: BMW ReachNow car-sharing service shuts down in Seattle and Portland following joint venture deal
University of Washington professor Kate Starbird talks about her research at a Wednesday night seminar. Credit: Ashley Stewart.
University of Washington professor Kate Starbird talks about social media research. Credit: Ashley Stewart.

Social media platforms are powerful resources during disasters, but important messages can easily be drowned out in 140 misinformed characters.

That’s why researchers at the University of Washington are working to develop an algorithm to identify rumors.

KSproject“We want to help people spot misinformation and communicate how it spreads,” researcher Kate Starbird told GeekWire. “We’re trying to come up with computational solutions that can identify rumors and when information is being corrected.”

Starbird is director of the UW’s Emergency Capacities of Mass Participation Laboratory, an arm of the Department of Human Centered Design and Engineering that examines massive interaction through social media and other platforms.

The former WNBA player-turned-associate professor and a team of students sifted through more than 10 million tweets from the Boston Marathon bombings, trying to identify and understand rumor-spreading behavior.

When two explosions near the finish-line killed three people and injured many more, emergency responders communicated through social media, but rumors and speculation overwhelmed useful information.

Few tweets in emergency events provide “information that is new and actionable and could help me make a better decision,” Starbird told a room full of students during a Wednesday night seminar. “The rest is pretty much noise and it can be hard to get through that noise to find a signal.”

Researchers collected seven days of tweets using Twitter’s Streaming API, which allows developers to collect data on search terms in real-time.

Boston2Using keywords “Boston,” “bomb,” “blast,” “explosion” and “marathon,” Starbird and her team collected samples, generated network graphs and coded tweets based on classifications, including misinformation or passing along a rumor, speculating and correcting or questioning a rumor.

The codes helped researchers identify how each rumor started and whether — or how often — they were corrected.

In one case, rumors spread of an 8-year-old girl who was killed while running in the marathon. Researchers identified the start of the rumor, how often it was passed along and found it accelerated when a photo of a young girl running a 5K started circulating with tweets.

The team also examined the patterns of other rumors, including those which incorrectly identified missing Brown student Sunil Tripathi as a suspect, claims that Navy Seals perpetuated the attacks for political reasons and stories of a woman who was killed in the bombings just before her boyfriend proposed.

Identifying the behaviors of rumors on social media could help people detect — and contain — misinformation in future disaster events, Starbird says.

“We’re still learning the rules of this new environment,” she told GeekWire. “We’re developing new norms and new understanding about what the best practices are for sharing information. People are learning lessons.”

Starbird’s research into social media rumoring during disaster events will continue with a two-year and nearly $500,000 grant from the National Science Foundation. Next, she plans to analyze, among other things, the credibility of certain nodes in a network.

Like what you're reading? Subscribe to GeekWire's free newsletters to catch every headline


Job Listings on GeekWork

Find more jobs on GeekWork. Employers, post a job here.