I guess it’s cool. It will be interesting to compare this with using fasttext as a baseline for classification.