Pro, Con, and Affinity Tagging of Product Reviews Todd Sullivan
The Task • Given a product review, what are the pros, cons, and affinities chosen by the reviewer?
Systems • Bag of words Naïve Bayes (baseline) • Maximum Entropy Classifier • Combinatorial Tag Optimization • Preprocessing Methods • Lowercasing, removing punctuation, stop words list, vocabulary restriction, etc. • Maximum Entropy Feature Selection • Combinatorial Tag OptimizationWeights Selection
Future Work • Human study of tagging performance • More advanced textual features using dependency trees / phrase structure trees • Better sentence boundary detection and tokenization (Balie did not work too well) • Focus on feature development for cons and affinities (spent all of our time on pros)