Context Aware Cyberbullying Classification Using Network Science

This research investigates the use of network science measures to enhance cyberbullying detection. Traditional detection algorithms rely on machine learning to classify content as ‘bullying’ based solely on textual analysis. However, these approaches often miss context, leading to incorrect classifications. Limited research has been done on incorporating network measures based on connections, to provide social context on platforms. This study integrates network features with textbased features, following a cyberbullying definition from the literature that includes aggression, repetition, harmful intent, power imbalance, and peer visibility. The results demonstrate that combining network context features and text features leads to higher accuracy in cyberbullying classification than methods relying only on text.