Visualization of Flicker Social Networks

GustavoG has recently posted a detailed analysis of the social network on FlickR.

To create this analysis he had to deal with a highly dense network, over 90,000 people from an initial seed of 1000 random FlickR users.

He describes his algorithm as:

what exactly is this "friend of a friend" approach?

It's an algorithm I made up today and I don't have a better name for it.
The network is too dense, too interconnected. I need to "trim it down" to be able to show something meaningful. So far, I had used a "mutual contacts" approach, i.e. if "A contacted B" and "B contacted A", the A-B relationship is mutual. If A has at least 50 such relations, it's included the mc-50 graph. This works fine, but I thought it might be obscuring some of the higher-order network structure.

In the "friend of a friend" approach, I first test every single contact relationship in the network, and decide whether to keep it or not. Let's say I'm testing the "A contacted B" link. To do this, I identify additional contacts in the form "A contacted C, and C contacted B". If I find at least 40 such "friend of a friend" links, I keep the "A contacted B" for the sc-40 graph. (The "sc" stands for "supported contact".) If not, the contact link is removed. After removing all such "unsupported" links, nodes not connected to anything else are removed from the graph.

Densely connected areas like the UAE cluster tend to remain largely unmodified. Rare links between such clusters tend to be dropped. This unravels the network into its better-connected components.