It's funny I was just pondering this one the other day. Twitter would be a smaller and well organized (relatively) data set, given the brevity of it's messages.

It's funny I was just pondering this one the other day. Twitter would be a smaller and well organized (relatively) data set, given the brevity of it's messages. Look for symbols, parse for a set of negative or positive markers in the remaining text. Add weight based on those markers and you could have a fairly accurate prediction engine. You could also track users over time, and their positive and negative indicators gauge for accuracy and add weights based on their record of picks. It would be a fairly simple system to write.