The New Scientist talks about "word bursts", which Jon Kleinberg at Cornell has been researching. Apparently, he's developed an algorithm to find bursts of words across documents, that might be used to identify trends in weblogs.

Dan Silverstein mentioned to me that while the specific technology might be really interesting, it's really hard to have a picture, or many overlapping pictures, of the entire Internet, in order to scan so many weblogs and find bursts of particular word activity, so this may take a while to solve. But one idea might be to target particular blogs by subject or reputation, in order to scan for particular word bursts.