Model-Agnostic Meta-Learning for Multilingual Hate Speech Detection

Hate speech in social media is a growing phenomenon, and detecting such toxic content has recently gained significant traction the research community. Existing studies have explored fine-tuning language models (LMs) to perform hate detection, these solutions yielded performance. However, most of are limited only English, neglecting bulk hateful that generated other languages, particularly low-resource languages. Developing classifier captures nuances with data extremely challenging. To fill gap, we propose HateMAML, model-agnostic meta-learning (MAML)-based framework effectively performs detection HateMAML utilizes self-supervision strategy overcome limitation scarcity produces better LM initialization for fast adaptation an unseen target (i.e., cross-lingual transfer) or datasets domain generalization). Extensive experiments conducted on five across eight different The results show outperforms state-of-the-art baselines by more than 3% cross-domain multilingual transfer setting. We also conduct ablation analyze characteristics HateMAML.