Every little piece of information you give away online can reveal something about you—but it seems your Facebook likes could reveal rather more than you bargained for.
New research by the University of Cambridge and Microsoft Research has been analyzing Facebook Likes to see how much information they really contain. After quietly—and, apparently, innocuously!—collecting data using Facebook apps like MyPersonality, the team of computer scientists crunched through it to see what they could find.
Turns out, they were able to determine things like gender, ethnicity, religion, political persuasion and more with over 80 percent accuracy. If you want, you can browse the entire data set yourself on a public wiki. The Wall Street Journal gives some specific examples:
As a measure of the computer model's accuracy, the researchers were able to distinguish between Democrats and Republicans in 85% of the cases; between black and white people in 95% of the cases; and between homosexual and heterosexual men in 88% of the cases.
The results were published yesterday in the Proceedings of the National Academy of Sciences. There are, as you might expect, some amazing quirks hidden within the data. Again, from the Wall Street Journal:
The researchers found... that "Likes" for Austin, Texas; "Big Momma" movies; and the statement "Relationships Should Be Between Two People Not the Whole Universe" were among a set of 10 choices that, combined, predicted drug use. Meanwhile, "Likes" for swimming, chocolate-chip cookie-dough ice cream and "Sliding On Floors with Your Socks On" were part of a pattern predicting that a person didn't use drugs.
If you're hoping that merely ignoring Facebook's Like button will shield you from such privacy intrusions—forget it. It's worth bearing in mind that this study is largely demonstrative, and not really tied to Facebook other than for its pool of data; in reality, every little digital breadcrumb you leave littering the internet could be used in the exact same way. Gulp. [PNAS via WSJ via Verge]
Image by Ksayer1 under Creative Commons license