Text this: Reversing uncertainty sampling to improve active learning schemes