Comment by svg7

It would be nice if you could report some accuracy metrics of this approach on well known text datasets, after hiding the label