Text classificationText classification is a fundamental task in machine learning, having useful applications in domains ranging from biomedicine to English literature. Text analyses are essential to tech mining, and the active learner presented here enables a range of technology analyses of big topical datasets not otherwise feasible. Many models, if not most, require hand-tagged data, which is expensive and time-consuming to produce. This constraint has recently given rise to a trend toward few-shot learning models, classifiers which require little training data. Because these models take so few examples, and machine learning output is only as good as its input data, it is essential that manual annotations be of the highest quality. To this end, I propose an active learningActive learning paradigm which selects documents to classify based on the goal of filling in “gaps” in the model’s understanding of the data. Compared to selecting documents randomly, this training workflow produces a higher average difference between the in- and out-of-category similarities of training documents for two corpora of biomedical text, as well as a lower number of classifications necessary to populate each category for the same. Such curation produces a robust training set while minimizing annotator error.