Active Learning with AutoNLP and Prodigy
Hugging Face Blog
Hugging Face's AutoNLP framework and Explosion's Prodigy annotation tool were combined to build an active learning pipeline for named entity recognition on BBC News articles. The researchers labeled approximately 250 samples iteratively, with model accuracy improving from 86% at 20 samples to 95.9% at 250 samples, while precision and recall reached 0.73 and 0.79 respectively. This approach eliminated the need for manual model selection, hyperparameter tuning, and preprocessing, allowing users without coding expertise to create labeled datasets and train production-ready transformer models.
