This was built by training a Bernoulli Naive Bayes classifer on a dataset of ~2k examples from movie reviews. I haven't tested the accuracy, as I know it's extremely low, although I believe if I preprocessed the data with word2vec to extract the proper features, I'd have a bit better results.
