ClusterPals

Use multi-dimensional clustering (see a comp sci geek for more information) to group people having similar tastes and tendencies. As an example, multi-dimensional clustering allows you to analyze thousands of shoppers and classify individuals according to whatever attributes turn out to be the primary identifying traits: single guys in their thirties who shop at night; minorites who always buy produce; old ladies who buy everything on sale.

The ClusterPals clusters would not be definable by any particular attribute (as in USENET newsgroups, where "fans of woody guthrie" might be the only qualifying attribute) but would contain people who are similar, to varying degrees, in a variety of areas. To find your cluster, you would answer a variety of questions much as you do for a Meyers-Briggs classification, except that Meyers-Briggs sorts you only along four known axes. The ClusterPals algorithm could create any number of clusters. There could be a process for adding/changing the interview questions. People could retake the interview & reclassify themselves from time to time if they wished.