Beyond Binary: Towards Embracing Complexities in Cyberbullying Detection & Intervention-A Position Paper

Cyberbullying (CB) has become a prevalent issue among children in the digital age. Social science research on CB indicates that these behaviours can manifest as early as primary school age and can have harmful and long-lasting effects, including an increased risk of self-harm. Drawing on insights from psychology, social sciences, and computational linguistics, this position paper highlights the complexity of CB incidents. These incidents are not limited to bullies and victims, but include bystanders with various roles, resulting in numerous sub-categories and variations of online harm. Despite the growing recognition of the complexities inherent in CB, existing computational approaches tend to oversimplify it as a binary classification task. They often rely on training datasets that may not comprehensively capture the full spectrum of CB behaviours. In addition to scrutinising the diversity of CB policies on online platforms and revealing inconsistencies in the definitions and categorising of CB-related online harms, this article also brings to attention the ethical concerns that arise when CB research involves children in role-playing CB incidents to curate datasets. This paper, through multi-disciplinary collaboration, seeks to address our position on strategies to consider while training or testing CB detection systems. Furthermore, it presents our standpoint on leveraging large language models (LLMs) like Claude-2 \& Llama2-Chat as an alternative to generate CB-related role-played datasets. By elucidating the current research gaps and presenting our standpoint, we aim to aid researchers, policymakers, and online platforms in making informed decisions regarding the automation of CB incident detection and intervention. By addressing these complexities, our research contributes to a more nuanced and effective approach to combating CB especially in young people.