What your Facebook 'likes' say about you
An algorithm can predict a user's race, religion, political affiliation — even cigarette, alcohol and drug use — based purely on their Facebook activity.
Mon, Mar 11, 2013 at 10:20 PM
Photo: Annette Shaff/Shutterstock
A new study shows that public Facebook "likes" reveal a person's private traits with spooky accuracy, meaning that even if Big Brother isn't watching, advertisers may be.
Michal Kosinski, a psychometrician at the University of Cambridge, and colleagues collected consenting Facebook users' personal traits from a questionnaire app called My Personality. Pairing that data with the users' publicly available "likes," Kosinski's team developed an algorithm that figured out which likes corresponded to which personality traits.
"We're able to predict personality with kind of mind-blowing accuracy," Kosinski said.
The algorithm could predict a user's race; religion; political affiliation; and cigarette, alcohol and drug use. But it could also predict "completely crazy things," Kosinski said.
For example, the software predicted with 60 percent accuracy whether a Facebook user's parents had divorced, because users with separated parents had a higher probability of "liking" the statement "preoccupied with relationships."
In many cases, the algorithm made connections that wouldn't be immediately obvious to a human. "If I looked at those likes, I wouldn't be able to say, 'Oh, this person is gay,'" Kosinski said. "But the computer is able to do it very easily." [How Facebook Sells Your Personal Information]
In fact, less than 5 percent of the users that the algorithm labeled as gay had liked explicitly gay-related topics. On the other hand, liking "Britney Spears" or "Desperate Housewives" was "moderately indicative" of being gay, as was liking the musical "WICKED."
This raises an interesting privacy conundrum, Kosinski said. "Everyone," from Facebook to Google to your credit card company to the advertisers that target you online, "is storing the same kind of data. I am convinced they never use this data to predict things like sexual orientation. They may predict what you want to buy … But the trouble is, a weird conservative government could do it, and then some people would be in serious trouble.
"I hope this paper, because I love Facebook, would start a discussion about how to solve this issue."
In the meantime, Kosinski is studying why certain likes correlate with certain personality traits. For example, users who liked "Sephora," "Harley Davidson" and "I Love Being a Mom" were less likely to be intelligent, whereas likes that predicted high intelligence included "Thunderstorms," "Science," "The Colbert Report" and "Curly Fries."
Wait, curly fries?
"I have no idea," Kosinski said. "This is something I'm studying at the moment."
Related on LiveScience and MNN: