I was talking with Greg and Max today about social networks, thinking of ways the Friendster model could solve all the worlds problems. We went down a few paths….

Media Recommendations:
I want to get into new music, new books, etc. Amazon has gotten me part of the way there, but I have a problem with their algorithms.. if I buy a jazz CD, I’ll see lots of jazz CDs in my recommendations… until I buy a polka CD, and then it’s all polka.

Here’s what I might want out of a good recommendations system:

- A human should make the recommendations by revealing their likes and dislikes to the system.

- Once you have these rankings, show me individuals within the network who have the most overlap with me. (Amazon says: people who bought X also bought Y, but that is averaged out and therefore not as interesting.. I don’t trust the averages as well as I might trust one person I like the taste of)

- Let me build up a trust relationship with other individuals based on how well I like what they recommend. The strength of these relationships will decide the likelihood that some new item of theirs will be recommended to me.

- Understand that music recommendations != book recommendations != electronics recommendations. I trust different people for different things.

What I think we have here is a mesh network. It’s sort of peer-to-peer instead of Amazon’s more centralized averaging system. Amazon may have attempted to do something more like this (something more “human”) by adding those recommendation lists, but the flaw is that I don’t want to know what some random dude who liked this one item I’m looking at also liked. I need more than one data point.

My current solution to the recommendations problem, at least for music (which I go through a lot of very quickly), is a simple mix CD swapping club. The other members just happened to have fairly congruent tastes with mine, and there were enough things I’d never heard of to keep me excited about it. We meet once every week or 2 and dicuss (for an hour!) a mix CD someone made. I’ve been around for 2 CDs worth so far and its opened me up to about 5 new artists I like that I’d never have heard of otherwise.

Amazon doesn’t do that for me.

Alumni Networks: Why isn’t Friendster-style networking sold to schools for alumni (or even current student) networks? Rather than providing a list of alumni who graduated in 2000, why not allow a Friendster network within the class. Then when I ask “Who was Agnus Beef? Where do I remember her from?”, I could see that I knew Agnus through my friend Chuck… etc.

Edward Tufte, in his recent PowerPoint slam, The Cognitive Style of PowerPoint, complained that a simple list of items is uninteresting because it provides no relationships between items… and this is one of the fundamental problems with PowerPoint’s list-based presentation model, he says. If you could have arrows between list elements, to show relationships, you’d add an entire dimension of information.

But I’m getting off topic. Time for sleep.