An integrated user matching framework for cross-platform cyberbullying detection

Cyberbullying has become a global issue across social media platforms, where individuals switch platforms to perform such activity to evade detection. This study introduced a Cross-Platform Cyberbullying Detection Algorithm (CPCD-Alg), a state-of-the-art algorithm that combines user behaviour, network analysis, linguistic patterns, and textual data analysis to identify cyberbullying within and across different social networks. Using two synthetic datasets denoting Platform A and Platform B, the proposed algorithm CPCD-Alg finds cyberbullying on a single platform by utilising machine learning and deep learning models, integrating BERT for textual analysis and graph-based methods for network analysis. After detecting cyberbullying on a single platform, the approach is then extended to cross-platform detection by matching features like behavioural analysis, linguistic patterns and network analysis. The approach does not consider username matching because users often use different names on different platforms to evade detection. The results show that adding multidimensional features enhances accuracy, with LSTM achieving 98% accuracy on single-platform text-based detection and provided graphs where network clusters were shown. For cross-platform, the system could reach an accuracy of 60%, reflecting the challenges imposed by matching user identities across platforms with synthetic datasets. The study developed a foundation for cross-platform cyberbullying detection by combining features other than only textual data as users could easily evade textual method detections. Future research will focus on extending the study to real-world datasets, to promote a safer environment for online users.