Multilingual Hate Speech Detection Using Meta-Transfer-Learning

—The fast expansion of online platforms created a record-setting rise in hate speech, undermining safe and wel coming online spaces. This project contributes a multilingual hate speech system based on meta-learning and transfer learning to achieve improved detection for languages. We employ a pre trained mBERT model for the English language and fine-tune it to the low-resource language of Hindi, Hinglish, Marathi and Bengali using meta-learning and transfer learning methods. With the incorporation of these practices, the system generalizes remarkably well to a new language given limited labeled exam ples, with impressive accuracy, scalability, and versatility. The proposed method greatly increases detection rates, reduces bias, and promotes healthy digital interactions ultimately leading to an improved safer online space. Besides, the framework of the system supports easy adaptation to other languages with small corpora.