How Facebook knows if you're going to break up before you do

It's really pretty elementary

Couple on computer
(Image credit: (Thinkstock))

If your relationship is Facebook official and you're still getting ads for JDate and Plenty of Fish in your news feed, the social networking site may have realized something you haven't: You and your significant other are headed for Heartbreak Hotel.

A new paper co-authored by Cornell University computer scientist Jon Kleinberg and Facebook engineer Lars Backstrom suggests Facebook data can not only predict whom you're in a relationship with, but if the relationship will go kaput.

Using the rather enormous sample size of 1.3 million Facebook users, Kleinberg and Backstrom set out to devise an algorithm for figuring out how to successfully determine if two people were in a relationship.

Subscribe to The Week

Escape your echo chamber. Get the facts behind the news, plus analysis from multiple perspectives.


Sign up for The Week's Free Newsletters

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

Sign up

It turns out the number of annoyingly cutesy selfies you post together or even the amount of friends you have in common aren't the prime predictors. In fact, what they discovered was that the best indicator of a relationship is how many of your mutual friends are not connected to each other.

This metric is called dispersion and it "measures mutual friends, but also friends from the further-flung reaches of a person's network neighborhood," says Steve Lohr at The New York Times.

Having high dispersion, or "when two people have widely dispersed clusters [of friends] that are linked mostly via the couple," writes Jim Edwards at Business Insider, is the best predictor of whether those two people are in a relationship. In fact, Kleinberg and Backstrom were able to correctly predict a user's spouse 60 percent of the time and a non-married partner one third of the time using this algorithm (a pure guess would produce 2 percent accuracy).

While cluster graphs and nodes of data may make it sound complicated, it's actually pretty intuitive when you consider the hallmarks of a strong relationship. Allison P. Davis at New York explains:

If you're friends with Joe, and share many mutual friends within the city you live in, and Joe is also friends with your childhood best friend, you and Joe are probably sleeping together. [New York]

This concept makes sense even without Facebook telling us. As Kleinberg tells The New York Times, "a spouse or romantic partner is a bridge between a person's different social worlds." They're the ones who go to annoying family events, office cocktail parties, and homecoming games at your alma mater. Basically, they meet and interact with a lot of your friends and relatives.

Of course, just as intuitively, it goes the other way for a bad relationship. Let's say Joe has no connection to your first cousin in Toledo, your bestie from sleep-away camp, your closest friend from study abroad; then you probably aren't in a particularly strong, long-term relationship. And when couples in a relationship don't have a high dispersion rate, they are 50 percent more likely to break up within the next two months.

Again, none of this is so shocking, except for the fact that Facebook has the frightening ability to detect all of these patterns before you take off the rose-colored glasses and realize that the Grand Theft Auto addict who doesn't take you to his uncle's birthday party or refuses to attend your high school reunion isn't your soul mate.

But who knows? With a little bit of time, Facebook may find that person for you, too.

To continue reading this article...
Continue reading this article and get limited website access each month.
Get unlimited website access, exclusive newsletters plus much more.
Cancel or pause at any time.
Already a subscriber to The Week?
Not sure which email you used for your subscription? Contact us