Advances in AI-Based Methods for Cyberbullying Detection on the X Platform: A Systematic Review

Abstract

Cyberbullying on social media platforms presents a serious and growing concern due to its psychological and social consequences, particularly among young users. The X platform, known for its high user activity and informal language, has become a prominent site for the spread of online abuse. In response, AI technologies, especially those based on ML, DL, and NLP, have been explored as automated solutions for detecting harmful content in real time. However, limitations persist in achieving consistent performance, particularly in multilingual environments and in the face of imbalanced data distributions. This systematic review investigates recent academic research focused on AI-based approaches for detecting cyberbullying on the X platform. Following the PRISMA framework, 30 peer-reviewed studies published between 2020 and 2025 were analyzed to identify trends in data collection, preprocessing, feature extraction, and model development. The review includes a comparative evaluation of traditional ML models (e.g., SVM, RF), neural network-based architectures (e.g., CNN, LSTM), and transformer-based models (e.g., BERT, XLM-R). The findings show a clear methodological shift toward transformer models, driven by their ability to handle informal, context-dependent language. Nonetheless, many challenges remain, including inconsistent dataset standards, lack of language diversity, and limited attention to fairness and model interpretability. This review provides a comprehensive overview of current research, highlights the strengths and limitations of existing methods, and offers a foundation for further development of reliable, ethical, and scalable cyberbullying detection systems for the X platform.