[HelenL14]

Maybe going beyond simple wordcount to determining statistical probabilities of certain words / phrases being mentioned together & at "high" frequencies would taper down the number of words you need a baseline for (because I also don't assume the baseline will be static or even near stable - changing trends of discussion topics)

This idea also ties into the "new news" idea a bit. What we've seen from events like #iranelection is that twitter is a good (but could be great) tool for detecting news at the onset of events (before big stations pick up on it), but really starts to get bogged down with "spam" or RT's once a topic gets picked up. I'd like to see news that integrates these abnormal spikes in tweets that hint at a bigger underlying story w/some sort of follow-up with more context about the event.