Many of the people here are clearly interested in the human side of social software: exploring how human relationships change when they're mediated through some technological means. Another interesting side of it, though, is what the machines can tell us about ourselves and our relationships -- information that we might not actually be aware of.

There are a few obvious ways that one might explore this: rather than depending solely upon the explicit, stated relationships between people, what happens if the machines that are mediating those relationships track interactions or types of behavior? Alice frequently initiates discussion threads, as does Dennis, but Bob and Carol both tend to respond to threads started by others. Edward frequently responds to Alice's postings, as does Carol (and perhaps Edward and Carol are responding to posts in different "tribes" and therefore missing the implied relationship between them). Edward and Carol don't have a stated relationship to one another, but is it significant that they interact with the same person? Dunno, but possibly an interesting direction to explore.

On a slightly different track, since latent semantic analysis is a big deal in some other areas of technology where I spend time poking around, I ended up wondering whether some of these principles could be applied.

There are a few interesting papers out there on "discriminative clustering of text documents," for example. By analyzing the contents of text documents, machines can rather accurately group them (the groupings generally end up being very similar to the groupings applied by human editors).

Note: this has been applied primarily to academic and scientific papers (so the texts are generally internally consistent and homogeneous), which leaves a potentially gaping hole in the speculation to follow. But...let's say that the body of text that someone contributes to a site such as this one constitutes their "text;" the machines then group these texts, creating meta-tribes of people who write about the same sorts of things.

This is, of course, potentially problematic as the texts in this scenario may contain so much "noise" that effective grouping is impossible. In addition, this analysis won't be able to factor context in, opening up the possibility for pretty weird classifications.

Still, it's an interesting area. We hear a lot about what you lose in machine-mediated social relationships -- are there things that can be enhanced or added by those machines?