As an extra-credit assignment you can try to do something smarter/better with than our NB approach. There are a lot of things you could do here. You might for example experiment with a better classifier (maxent, svm, random forest, etc.) Or you might do some feature selection to get a better set of features for NB. The only stipulation is that the task stays the same.
Note that since the NB performance is quite high you will have difficulty getting a big improvement here on accuracy. Other improvements might involve creating smaller models that do just as well. Or using vastly smaller amounts of training data to get the same effect.
I'll provide an additional test set for this part of the assignment.