Data expansion using back translation and paraphrasing for hate speech detection

Abstract With proliferation of user generated contents in social media platforms, establishing mechanisms to automatically identify toxic and abusive content becomes a prime concern for regulators, researchers, society. Keeping the balance between freedom speech respecting each other dignity is major platform regulators. Although, automatic detection offensive using deep learning approaches seems provide encouraging results, training learning-based models requires large amounts high-quality labeled data, which often missing. In this regard, we present paper new method that fuses Back Translation method, Paraphrasing technique data augmentation. Our pipeline investigates different word-embedding-based architectures classification hate speech. The back translation relies on an encoder–decoder architecture pre-trained corpus mostly used machine translation. addition, paraphrasing exploits transformer model mixture experts generate diverse paraphrases. Finally, LSTM, CNN are compared seek enhanced results. We evaluate our proposal five publicly available datasets; namely, AskFm corpus, Formspring dataset, Warner Waseem Olid, Wikipedia comments dataset. performance together with comparison some related state-of-art results demonstrate effectiveness soundness proposal.