I've recently been working with Caroline Meeks to deploy and run an extranet that features social bookmarking as a killer app. It's truly amazing what Caroline, who is CEO of Solution Grove, and her colleagues there have been able to do with limited time and resources, using .LRN and OpenACS as the substrates for our efforts.
Caroline approached me last week with an idea to take this even further by "graphing" tag patterns by users in order to calculate what I'll call a "proximity of interests" between any two users. In a world where filtering information has gotten much more challenging than finding it, this could be very useful.
Here's how it would work. The proximity between me and any other user in the system can be calculated according to the frequency with which another person uses tags I also use. This happens on two dimensions -- one is how often another person has used each tag I've used; another is the overlap between the tags we use, regardless of tagging frequency. So if a person is in my "high-high quadrant" (we're both using the same tags a lot), maybe I'd be interested in subscribing to the other person's bookmark stream. This might be especially true if "a lot" is defined absolutely for the other person, if only relatively for me, so I can filter out the great Long-Tail unwashed. "Show me people who have used my top ten tags 100 times or more."
A smart system would crank out this proximity score every now and then, and suggest folks I might want to drop and new ones I might pick up, based on some thresholds I or it might provide.
While we figure out if and how we might want to pursue this idea, you might want to get in touch with Caroline if it sounds interesting to you. Or, if you've seen this done elsewhere, let us know! We've seen systems that calculate network proximity based on explicitly-completed profiles (including some that use tags to help complete the profile), but I've seen none that suggest connections based on their actual interactions with the system.
Postscript: "Graphic Friendships, Part II"