Real-Time Multi-Lingual Hate and Offensive Speech Detection in Social Networks Using Meta-Learning

A rapid increase in users on social media has given rise to a vast amount of user-generated content, including hate speech and offensive language. Such content can have serious negative consequences, ranging from psychological harm to inciting violence and discrimination. Existing studies have explored different deep learning and Natural language processing (NLP) methods to perform hate speech detection, and these solutions have yielded significant performance. Most existing solutions are limited to detecting hate speech only in English with less focus on content generated in other languages, particularly in low-resource or regional languages. The goal of this paper is to address this challenge of hate speech detection for low-resource languages and propose a tool that could provide a real-time prediction for social media posts. In this study, the main focus was on English, Hindi, Hinglish, Bengali, and Marathi languages which are commonly used in social media platforms in India. A meta-learning-based model was employed to perform hate speech detection in these languages. The proposed method helps to overcome the limitation of data scarcity and provides fast adaptation to an unseen target language. Extensive experiments were conducted on datasets comprised of different regional languages spoken in India. Accuracy, Precision, recall, and F1-score metrics are used to evaluate the model's performance. The results show that when the dataset size is small, meta-learning-based models perform better than traditional fine-tuned language models.