Deep Learning-Based Detection of Hate Speech and Toxic Content in Nigerian Social Media

Source abstract

The use of social media in Nigeria for communication, information access, and public discussion cannot be overemphasized. However, some individuals and organizations exploit it negatively, spreading hate speech, misinformation, and toxic content. Such abusive online behavior has led to politically, socially, and ethnically motivated protests across Nigeria. Manual moderation, where users flag abusive posts or comments, is largely ineffective due to the high volume of content generated online. Although deep learning has been widely applied to address this problem, most models are primarily trained in English. This limits performance in contexts such as Nigeria, where a significant portion of social media communication occurs in Nigerian Pidgin. This study presents a deep learning-based framework to automatically identify harmful content in Nigerian Pidgin. Leveraging models such as LSTM and Transformers, the framework captures contextual and linguistic patterns, including informal expressions and slang. Preprocessing strategies are tailored to local language nuances and cultural contexts. Conceptual evaluation indicates that this framework can support real-time content moderation, enhance online safety, and serve as a foundation for further research on AI-driven social media governance in Nigeria.