Comment by snikolov

I agree with you --- this seems like a perfectly good way to tackle the problem of trend prediction directly. What we had in mind was something that would be more generally applicable to any kind of time series data, and we figured it would be interesting to test it on Twitter trends.