As long as you have an appropriate training set, I don't see why it would be significantly different. You might run into problems if you train the classifier on

As long as you have an appropriate training set, I don't see why it would be significantly different. You might run into problems if you train the classifier on some radically different dataset but then try to run it against twitter accounts.