Ross Mayfield writes about some nifty work by Valdis Krebs to map the network of relationships at Ryze, an online business networking group, and the weblog tribe on Ryze.
Here’s some more analysis that would be really interesting:
a) identify clusters of blogs — blogs that share a number of blogroll blogs in common (first filter out the most popular blogs).
b) use text analysis to identify the topics the clusters have in common by identifying words they use much more frequently than average.
This would identify groups of New York bloggers, Java bloggers, warbloggers, etc. Groups wouldn’t be mutually exclusive; lots of people would belong to more than one cluster.
Blogrolling.com exports blogrolls in RSS and OPML format, so that might be a workable dataset. They have 6915 blogrolls with 108278 links.
The math to do this is pretty far over my head — so this one goes out to the Lazyweb!
Adina — one of the classic tools for cluster analysis results in something called a “dendogram”, see http://trochim.human.cornell.edu/tutorial/flynn/cluster.htm for a very small sample of it. In short you compute a distance metric between node pairs and then compute a spanning tree in that space. The algorithms are not hard to come up with *something*, but getting something *meaningful* is very hard and a job for humans.