Comment by ameasure

Fascinating paper but the benchmarks seem incredibly weak. 5000 features for a bag of words model is nothing,these models normally have tens or hundreds of thousands of features.