[liuhenry]

From the linked paper [1], it looks like the scores are calculated based on a concept of "dispersion", which is how closely a particular individual in your network is connected to each of the separate clusters of people you know.

Each of the clusters in the graph represents a "social foci", such as co-workers, high school classmates, college classmates, etc.

The dispersion score calculates how well someone you know is connected to these multiple disjoint clusters of people in your life, and tries to show that the person with the highest score is most likely to be your romantic partner.

This probably doesn't work so well in trying to identify friends, since it's likely that a close friend from college may not know any of your friends from work, or your high school classmates.

1 - http://arxiv.org/pdf/1310.6753v1.pdf