I signed up for the beta of Eurekster, which uses a collaborative filtering approach to web searching, based on the searches of your 'friends' in the Eurekster network.

"What's Hot in Your Network

eurekster uses the six-degrees of separation concept to learn from your extended network of contacts and deliver you prioritised results based on the success and proximity of the searches they have done."

I can imagine that it is helpful to know "what's hot" in your network (a la Blogdex), and if your network is homogeneous (and hard-working in training the system) you will get better search results than keyword- or editorial-based solutions.

My problem is that my network is heterogeneous: really a collection of independent networks. As more and more of my networks are brought into Eurekster, the group will more and more approximate a random sample of people, and this will cancel out the social network effect. The answer is that I really need to be able to partition the network into discrete subnetworks: what are my social software buddies looking at today? What about my personal friends? What about people in the 20194 area? Until social networks attack this angle, we will be dealing with a very coarse-grained approximation for what is actually going on in social interactions.