Cs123a Term Project: Classifying Yelp Data

This paper describes our process of building a classifier for ratings data as assigned in spring semester of Machine Learning (CS123A), taught by Pengyu Hong at Brandeis University. The data set can be characterized by a large number of sparsely populated attributes (n=291) corresponding to ratings of 1 through 5 from the online ratings system, Yelp. To attack this problem we tried a wide variety of algorithms and compared their results, concluding that the Logistic Regression and SMO algorithms provided the most robust classifications, though neither were able to break the 50% accuray barrier. We also discuss a failed attempt at employing a 2-step approach to classification that we believe may yield better results with further experimentation.