A Hate Speech Detection Approach Using Transfer Learning with Multiple Idioms

Nowadays, there is a growing connection among individuals promoted by the Internet, which provides opportunities for expressing their viewpoints through social media platforms. However, this expanded freedom of expression has been susceptible to the propagation of hate speech, a phenomenon that can precipitate unlawful conduct and potentially engender detrimental psychological ramifications. In response, computational technology has emerged as a valuable tool for identifying and mitigating hate speech on social media. In this chapter, we used five datasets to detect hate speech related to politics on social media. These datasets encompass the English, Italian, Filipino, German, and Turkish languages. In pursuit of hate speech detection, our study advocates adopting a Pre-Trained Language Model (PTLM) with Cross-Lingual Learning (CLL). We tried to detect hate speech in two languages (English and Italian) using English BERT and Italian BERT. We used Zero-Shot (ZST), Joint Learning (JL), Cascade Learning (CL), JL/CL, and CL/JL+ approaches. These techniques demonstrated efficacy in detecting hate speech. We obtained 94.8% in the F-score metric using English BERT and 93% using Italian BERT.