ToxVis: Enabling Interpretability of Implicit vs. Explicit Toxicity Detection Models with Interactive Visualization

Uma GunturiXiaohan DingEugenia H. Rho

Mar 2023

摘要

The rise of hate speech on online platforms has led to an urgent need foreffective content moderation. However, the subjective and multi-faceted natureof hateful online content, including implicit hate speech, poses significantchallenges to human moderators and content moderation systems. To address thisissue, we developed ToxVis, a visually interactive and explainable tool forclassifying hate speech into three categories: implicit, explicit, andnon-hateful. We fine-tuned two transformer-based models using RoBERTa, XLNET,and GPT-3 and used deep learning interpretation techniques to provideexplanations for the classification results. ToxVis enables users to inputpotentially hateful text and receive a classification result along with avisual explanation of which words contributed most to the decision. By makingthe classification process explainable, ToxVis provides a valuable tool forunderstanding the nuances of hateful content and supporting more effectivecontent moderation. Our research contributes to the growing body of work aimedat mitigating the harms caused by online hate speech and demonstrates thepotential for combining state-of-the-art natural language processing modelswith interpretable deep learning techniques to address this critical issue.Finally, ToxVis can serve as a resource for content moderators, social mediaplatforms, and researchers working to combat the spread of hate speech online.

展开全部

图表提取