Other Insights

• Similar performance achieved using news stories with titles harvested from the web

• EM instead of Naïve Bayes is slightly better

• Small degradation for corrupted text (speech recognition, machine translation)

Previous slide

Next slide

Back to first slide

View graphic version