It's at the lower level character rather than word level which is unique. The convolutional net also doesn't need to be told what each word's role is, but rather learns that feature itself.
What surprises me is that (BOW model + logistic regression) works just fine in most of the benchmarks(except for Amazon Review), interesting paper anyway. Could it be that because the vocabulary for BOW is limited to 5000, a lot of information is lost?
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.
[+] [-] sgt101|11 years ago|reply
[+] [-] bradneuberg|11 years ago|reply
[+] [-] eva1984|11 years ago|reply
[+] [-] ameasure|11 years ago|reply
[+] [-] sushirain|11 years ago|reply
* Compare to RNNs with character level input.
* Compare to dedicated methods of sentiment analysis and topic categorization.
[+] [-] petercooper|11 years ago|reply