Abstract
Detecting hate speech in videos remains challenging due to the complexity of multimodal content and the lack of fine-grained annotations in existing datasets. We present HateClipSeg, a large-scale multimodal dataset with both video-level and segment-level annotations, comprising over 11,714 segments labeled as Normal or across five Offensive categories: Hateful, Insulting, Sexual, Violence, Self-Harm, along with explicit target victim labels. Our three-stage annotation process yields high inter-annotator agreement (Krippendorff's alpha = 0.817). We propose three tasks to benchmark performance: (1) Trimmed Hateful Video Classification, (2) Temporal Hateful Video Localization, and (3) Online Hateful Video Classification. Results highlight substantial gaps in current models, emphasizing the need for more sophisticated multimodal and temporally aware approaches. The HateClipSeg dataset are publicly available at this https URL.
Abstract (translated)
在视频中检测仇恨言论仍然是一个挑战，这主要是由于多模态内容的复杂性和现有数据集中缺乏精细标注。我们介绍了HateClipSeg，这是一个大规模的多模态数据集，其中包含视频级别和片段级别的注释，包括超过11,714个被标记为正常或五个 Offensive 类别中的一个：仇恨、侮辱、性、暴力、自残，并且带有明确的目标受害者的标签。我们采用三阶段标注过程以获得高一致性（Krippendorff 的 α = 0.817）。为了评估性能，我们提出了三项任务：(1) Trimmed Hateful Video Classification (修剪后的仇恨视频分类)，(2) Temporal Hateful Video Localization (时间轴上的仇恨视频定位)，以及 (3) Online Hateful Video Classification (在线仇恨视频分类)。研究结果突显了现有模型在当前性能方面的显著差距，强调了需要更复杂的多模态和时间感知方法。HateClipSeg 数据集可以在以下网址公开获取：[此 https URL]（请将 [此 https URL] 替换为实际链接）。
URL
https://arxiv.org/abs/2508.01712
