So my question is how Supervised FastText works for the most part. I understood in the original paper they use bag of n-grams for features, but then they released a paper with enriching the word vectors using character-n-grams. Using the FastText package released by FAIR, when learning a supervised model you incorporate both a character-n-gram size and word-n-gram size, so I am a bit confused on what they do for the implmentation?

Do they use the character-n-gram to learn representation as its own objective and then use the n-gram BOW ontop of the learned representation?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.