References & Citations
Computer Science > Computation and Language
Title: A Holistic Approach to Undesired Content Detection in the Real World
(Submitted on 5 Aug 2022 (v1), last revised 14 Feb 2023 (this version, v2))
Abstract: 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.
Submission history
From: Todor Markov [view email][v1] Fri, 5 Aug 2022 16:47:23 GMT (7299kb,D)
[v2] Tue, 14 Feb 2023 19:39:32 GMT (7300kb,D)
Link back to: arXiv, form interface, contact.