Davidson et al. (Twitter tweets)

该数据集包含了从8500万条推文中筛选出的2.5万条包含Hatebase词汇的推文。其中，CrowdFlower的工作人员手动将这2.5万条推文中的24,802条划分为三个类别：仇恨言论、具有攻击性但非仇恨言论，以及既不攻击性也不属于仇恨言论。该任务的目的是进行仇恨言论分类。

This dataset includes 25,000 tweets containing Hatebase vocabulary, filtered from 85 million original tweets. Among these 25,000 tweets, 24,802 were manually categorized by CrowdFlower staff into three classes: hate speech, offensive but non-hate speech, and neither offensive nor hate speech. The objective of this task is hate speech classification.