Hate Speech Detection using Deep Neural Networks

This study includes a comprehensive study on detecting the hateful content in two benchmark datasets using standard sequence models such as RNN, LSTM, Bi-LSTM along with their stacked models, hierarchical attention hybrid neural networks, and capsule networks. ETHOS binary dataset which consists of 998 tokens from YouTube and Reddit comments and HateXplain multi-class dataset in which the corpus containing 20K tokens crawled from Twitter and Gab annotated with three classes namely Hate, Offensive and Normal, annotated by Amazon mechanical Turk workers has been used along with fastText word embeddings. Both datasets are publicly released in 2020 and 2021 respectively.