franciellevargas/HateBR

HateBR是一个专门针对巴西葡萄牙语仇恨言论检测的大规模、专家标注的数据集。该数据集包含了7000份来自巴西Instagram政治评论的文档，这些文档被专家手动标注，分为三个不同的层次：二元分类（攻击性与非攻击性评论）、攻击性级别（高度、中度、轻微攻击性信息）和仇恨言论目标。每个评论都由三位专家标注，确保了标注结果的高一致性。

HateBR is a large-scale, expert-annotated dataset specifically designed for Brazilian Portuguese hate speech detection. The dataset comprises 7,000 documents from Brazilian Instagram comments made by politicians, manually annotated by experts across three distinct layers: binary classification (offensive vs. non-offensive comments), offensiveness level (highly, moderately, and slightly offensive messages), and hate speech targets. Each comment is annotated by three expert annotators to ensure a high level of inter-annotator agreement.