CEDL Hate Speech Detection Benchmark Dataset

It is derived from three publicly available benchmark datasets: Implicit Hate Corpus (IHC), Hateful Discussions (HD), and DynaHate, and reorganized according to the experimental settings of this study.The IHC dataset contains implicit hate speech samples collected from Twitter and is used for model training and in-domain evaluation.

The HD dataset consists of annotated discussion comments from Reddit communities and is employed to assess model performance under more recent data distributions. The DynaHate dataset is used as a cross-dataset benchmark to evaluate transferability and generalization under distribution shifts.The final dataset includes training, testing, and cross-dataset evaluation subsets with hate and non-hate labels. It can support research on hate speech detection, implicit hate recognition, knowledge distillation, and large language model-assisted text classification.