ToxCCIn: Toxic Content Classification with Interpretability
This addresses the need for interpretability in toxicity detection on social media, where manual appeals require clear explanations, though it is incremental as it builds on existing transformer models.
The paper tackled the problem of opaque decision-making in transformer models for toxic content classification by proposing a method that assumes a post's toxicity is at least as high as its most toxic span, resulting in explanations that exceeded the quality of Logistic Regression in a human study.
Despite the recent successes of transformer-based models in terms of effectiveness on a variety of tasks, their decisions often remain opaque to humans. Explanations are particularly important for tasks like offensive language or toxicity detection on social media because a manual appeal process is often in place to dispute automatically flagged content. In this work, we propose a technique to improve the interpretability of these models, based on a simple and powerful assumption: a post is at least as toxic as its most toxic span. We incorporate this assumption into transformer models by scoring a post based on the maximum toxicity of its spans and augmenting the training process to identify correct spans. We find this approach effective and can produce explanations that exceed the quality of those provided by Logistic Regression analysis (often regarded as a highly-interpretable model), according to a human study.