The study, which will be presented next year, studied a random sample of 103 million Facebook users who listed themselves as in a romantic relationship. according to Quartz. This gave Backstrom and Kleinberg billions of links and millions of nodes (points where links overlap) to work with.
Rather than study the conventional notion of “embeddedness,” or how many friends in common two people had, the authors looked at “dispersion.” Put simply, dispersion looks at connections in separate circles of friends.
What this means is that one half of a couple may not be familiar with his or her spouse’s close friends, but would would be involved in other circles of relations including, for example, distant family.
The study, in short, proved that the higher the levels of dispersion, the more likely two people were in a relationship, and the more likely the couples were to stay together. Therefore, the more of your friends you introduce your partner to, the more likely the two of you are to stay together.
When the authors didn’t know the relationship statuses of their random samples, prediction rates dropped somewhat, but an analysis of a random sample of 129,000 still predicted relationship statuses 60 percent of the time.
Mashable reports that Facebook could use this data to determine the most relevant content for one’s news feed.
“If we can do a better job of identifying all the most important people in your life, there is a lot of opportunity to make Facebook better,” Backstrom said.
Previous studies have shown that this isn’t the first time data analysis on Facebook could be used as a prediction method. Earlier this year, a study of 58,000 US Facebook users showed that it could figure out like race, sexuality, substance abuse and more just by looking at Facebook “likes.”
