Raising the baseline for high-precision text classifiers