Possible? Yes. Practical? I'm not a genius programmer (just an ordinary one), but the only ways I can think of doing it would be very labor intensive as it would involve hand-censoring tweets. If you were able to come up with an algorithm that is able to read positive tweets from negative ones, then you'd be golden; that sort of language detection is definitely high on the list of things to be able to do with a computer. I don't think we're to a point where it's practical or effective enough yet, though, but it is definitely coming.