We present a holistic approach to building a robust and useful natural language classification system for real-world content moderation. The success of such a system relies on a chain of carefully designed and executed steps, including the design of content taxonomies and labeling instructions, data quality control, an active learning pipeline to capture rare events, and a variety of methods to make the model robust and to avoid overfitting. Our moderation system is trained to detect a broad set of categories of undesired content, including sexual content, hateful content, violence, self-harm, and harassment. This approach generalizes to a wide range of different content taxonomies and can be used to create high-quality content classifiers that outperform off-the-shelf models.
@article{arxiv.2208.03274,
title = {A Holistic Approach to Undesired Content Detection in the Real World},
author = {Todor Markov and Chong Zhang and Sandhini Agarwal and Tyna Eloundou and Teddy Lee and Steven Adler and Angela Jiang and Lilian Weng},
journal= {arXiv preprint arXiv:2208.03274},
year = {2023}
}