Auto-Categorization of Businesses on Yelp.com

We built a system to infer semantic categories of businesses (Auto-Categorization) from Yelp.com based on business titles and reviews. Auto-Categorization is accomplished by training a Näıve Bayes classifier on labeled word frequencies (supervised learning). We extend this classifier by using Expectation-Maximization to bootstrap off of unlabeled word frequencies (semi-supervised learning). Given more time, we would have liked to look at adding named entities and chunking of words with high mutual information as additional features to our model.