Deep learning models for culturally aware cyberbullying detection in Muslim societies: a systematic review

Cyberbullying, a form of harassment conducted online, poses significant risks to the mental health and wellbeing of individuals. While cyberbullying detection models have improved over time, they often overlook cultural and religious factors, particularly those relevant to Muslim communities. This paper provides a comprehensive review of deep learning models used for detecting cyberbullying, with a focus on their limitations in recognizing culturally specific forms of abuse within Muslim societies. Current models, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transformer-based architectures, have shown promise in detecting general cyberbullying, but they often fail to account for the nuances of Islamophobic hate speech, religious symbolism, and culturally sensitive content. We identify significant gaps in existing research, including the lack of culturally enriched embeddings, limited labeled datasets from Muslim-majority regions, and challenges posed by multilingual and code-mixed content. Additionally, we highlight the need for incorporating sociolinguistic and religious features into the training of these models. We also emphasize the importance of developing multimodal detection systems that can process both textual and visual data, which are increasingly used in cyberbullying in general and are particularly relevant in Muslim contexts where images and memes often accompany text-based abuse. Finally, we outline future research directions aimed at creating more culturally aware, efficient, and accurate deep learning frameworks for cyberbullying detection, ensuring better protection for Muslim communities online.