3 ms·
Distributed Representations of Sentences and Documents
- PaulHoule 8y agoI like this idea much better than I like word vectors. Although word vectors are well-established, at the end of the day people want to classify documents, not classify word vectors. Also note that: (1) Sentiment analysis is where BoW goes to die (e.g. "not good" is roughly equal to "bad") and (2) I have not seen so many "beyond BoW" tasks other than sentiment analysis that have been well documented.