On Saturday night, George Zimmerman was found not guilty of second-degree murder in the death of Trayvon Martin. When the controversial verdict came down, some took to social media to express their shock and anger. Twitter quickly came to resemble a digital lynch mob, with some threatening to kill Zimmerman, while thousands took to the streets in protest across the country.
Many of the gatherings were peaceful, but we wondered: Is it possible to use social media to predict and prevent violent or riotous behavior? “We’re not there yet,” says Haile Owusu, director of research at SocialFlow, “and one has to wonder if it’s even possible to do that.”
Haile looked at some of the social media data that emerged following the Zimmerman ruling. “There’s a violent tone to a lot of what is written about Trayvon and Zimmerman that hasn’t precipitated anything,” he said. And in a nutshell, that’s why using social media to predict violent social uprisings is almost impossible. Saying is very different from doing, and social media is often used as a place to vent, and nothing more. Twitter cannot predict if or when violent words will become violent actions.
Another problem? At least in the U.S., we don’t have riots very frequently. “In the age of Twitter, there aren’t a lot of things that can be generally called riots,” says Haile. This results in a lack of data, which makes riots very hard to study. “There’s very few data points to substantiate a meaningful probability.”
Also, people who really plan to commit violent crimes aren’t likely to tweet about it. “There is a huge disincentive to use social media to demonstrate intent,” Haile says. “It will get you convicted if there’s a correlation between what you’re saying and what you do.” So while a handful of people who intend to commit a crime may broadcast their plans on social media, there are more people who won’t.
What if we could measure a geographical area’s emotional state? Sune Lehmann, an Associate Professor at DTU Informatics, Technical University of Denmark, and his team used Twitter data to do this. By assigning numerical values to certain words in tweets, both positive and negative, they were able to measure the average mood of specific states. But even this method isn’t able to truly predict violent behavior. “Clearly that’s not how it works, right? We’re doing some kind of crazy simplification,” Lehmann says.
What if you set a threshold so that, when the threshold was hit, it signaled that a riot was imminent? If enough people are tweeting with a certain keyword or hashtag, could that be a signal? “Most of the time it would be a false positive,” says Lehmann. “Conditions may be met, says the system, and then nothing actually happens.”
For now, Twitter is best for doing two things: spreading a message and monitoring events. But using social media as an oracle for mass instances of violence, while an interesting idea, is still pretty far off.
[Image: Flickr user Nesster]