Personality and emotion based cyberbullying detection on YouTube using ensemble classifiers

Abstract

ABSTRACT This paper investigates the effect of users’ personality traits and emotions expressed through textual communications on YouTube to detect cyberbullying using a series of ensemble classifiers. Personality traits were determined using the Big Five model whereas emotions were based on Ekman’s basic emotion theory. Annotated YouTube textual comments in English (N = 5152; i.e. 2576 number of bullying versus 2576 non-bullying instances) were used to detect cyberbullying incidents using several ensemble classifiers, including Random Forest and AdaBoost. Performance metrics revealed both personality traits and emotion to significantly improve the identification of cyberbullying presence, with accuracy and F-score values of more than 95%. Further fine-grained analysis revealed anger and openness to be more profound compared to other emotions and personalities. Further, neurotic individuals tend to be driven to cyberbullying by joy, disgust and fear. The findings show that personality and emotions play pertinent roles in cyberbullying, and the identification of specific traits and emotions can help in designing a more strategic intervention programme.