Bullying discourse on Twitter: An examination of bully-related tweets using supervised machine learning

Prior research shows that combining social science with big data can advance our understanding of key social issues like bullying. The current study examines the sharing and disclosure of bullying experiences through the use of Twitter data by including keywords that capture both face-to-face and cyberbullying experiences. Using human coded tweets and supervised machine learning, the study considers the role of the author in bullyingrelated tweets, identifies different types of bullying, analyzes why someone would share a bullying episode on Twitter, and examines the temporal patterns of bullying-related tweets. The study analyzed 847,548 tweets collected between August 7, 2019, and March 31, 2020. The results revealed that most of the tweets were shared from the perspective of the victim, included both general and online bullying, and the most common reason for posting was to report or to self-disclose. Bullying-related tweets were significantly longer than the average tweet and high profile incidents prompted an increase in posts. The results suggest that while Twitter may be a venue for bullying, it is also a space where users can find cathartic discussion and support. This study highlights ways that researchers, educators, and policymakers can utilize Twitter as a medium for positive change and harness machine learning to inform policy and anti-bullying initiatives.