To test and train our SVMs, we used a movie review dataset compiled by Pang and Lee (see Related Papers) found here. The dataset consisted of 1000 positive and negative movie reviews collected from iMDB. The files were first processed with Adwait Ratnaparkhi's sentence boundary detector MXTERMINATOR, which placed commas at the end of sentences. The dataset was then divided into ten-parts and a ten-fold cross-validation was performed. The learner was evaluated based on the percentage of test movie reviews (100 per fold of validation) that it classified correctly.