You could use the google predict api to to language analysis, ex this is a happy tweet, this is a sad one, etc. I bet there's someone on the team playing with just such implementations. https://cloud.google.com/prediction/

I mean, what would you do if you had access to over a trillion searches?