The Use of Unlabeled Data versus Labeled Data for Stopping Active Learning for Text Classification