November 3, 2013

Cornell Study: Facebook Predicts If Couples Will Break Up

Print More


Thanks to research conducted by Cornell Prof. Jon Kleinberg, computer science, and Facebook employee Lars Backstrom ’04 Ph.D. ’09, an algorithm can now recognize with greater accuracy whether a couple is likely to split up.

Kleinberg and Backstrom began conducting research in June 2011, seeking a way to utilize people’s Facebook friend networks and determine how to “better prioritize what they’re seeing in their newsfeed,” Kleinberg said.

What started out as an attempt to improve the relevance of Facebook users’ news feeds developed into a study of how to identify strong relationships — specifically romantic partners.

“If I just gave you the connections among a person’s friends, could you recognize their spouse or romantic partner?” Kleinberg said.

To find an answer, Backstrom and Kleinberg studied a sample of about 1.3 million anonymous Facebook users. The group consisted of people between the ages of 20 and 50 who had between 50 and 2,000 friends on Facebook.

Each of the 1.3 million individuals studied had declared their relationship status on their Facebook profiles to be “in a relationship,” “married” or “engaged,” according to Kleinberg and Backstrom’s research paper.

Kleinberg and Backstrom first examined how greatly friends were embedded in a friend network by looking at the number of shared mutual friends people had, Kleinberg said.

The two researchers found that the ability to determine someone’s romantic partner is not about how embedded a romantic partner is within your social circles, but about how dispersed their friendships are within your network.

For example, Kleinberg said, a Facebook friend with many common connections among your different social circles –– such as high school friends, college friends and work friends –– is more likely to be in a romantic relationship with you.

Kleinberg and Backstrom next looked at what implications dispersion theory has for relationship longevity.

Kleinberg said that when the algorithm incorrectly guesses the romantic partner, “there was a 50-percent greater likelihood that the relationship would actually be over, at least on Facebook, in another two months.”

Kleinberg and Backstrom’s research has been met with mixed student reactions.

One student said the research might help identify characteristics that make relationships last longer.

“I think there are certain benefits of knowing how social ties can predict the characteristics of the relationship. Maybe these facts can help lengthen a relationship,” Said Israilov ’14 said.

But Ross Tannenbaum ’17 said he thinks the research is incomplete because it does not take other aspects of relationships into account.

“You have to take into account when the relationship started, and also who ‘friended’ whom,” Tannenbaum said, adding that “the failure of the algorithm really doesn’t say much, for the strength of a relationship cannot be determined based solely on Facebook friends.”

Kay Xiao ’16 said that the findings elicited mixed emotions for her.

“I think it’s kind of cool and a little creepy at the same time. It’s a kind of off-putting that an algorithm can predict the likelihood that a couple will break up,” Xiao said.

Although Kleinberg acknowledged that the algorithm is limited because it cannot predict how long a relationship will last, he noted that the algorithm can still indicate whether there is an increased risk that someone will break up with their partner soon.

Kleinberg said that the research findings highlight a simple principle.

“If you have a romantic partner, and you’re not friends with a bunch of people from their social circle … does that say something deeper about what’s going on in the relationship?,” Kleinberg said.