Cyberbullying detection on twitter using Big Five and Dark Triad features. (15th April 2019)

• Record Type: Journal Article
• Authors: Balakrishnan, Vimala; Khan, Shahzaib; Fernandez, Terence; Arabnia, Hamid R.
• Is Part Of: Personality and individual differences. Volume 141(2019)
• Page Start: 252
• Page End: 257
• Publication Date: 2019-04-15
• Subjects: Cyberbullying -- Personality -- Dark triad -- Big five -- Twitter -- Random Forest; Personality -- Periodicals; Individuality -- Periodicals; Individuality -- Periodicals; Personality Development -- Periodicals; Personnalité -- Périodiques; Individualité -- Périodiques; 155.205
• DOI: 10.1016/j.paid.2019.01.024
• Languages: English
• ISSNs: 0191-8869

Abstract

Abstract: This paper presents a cyberbullying detection model based on user personality, determined by the Big Five and Dark Triad models. The model aims to recognize bullying patterns among Twitter communities, based on relationships between personality traits and cyberbullying. Random Forest, a well-known machine-learning algorithm was used for cyberbullying classification (i.e. aggressor, spammer, bully and normal), applied in conjunction with a baseline algorithm encompassing seven Twitter features (i.e. number of mentions, number of followers and following, popularity, favorite count, status count and number of hash tags). Findings indicate that factoring user's personality greatly improves cyberbullying detection mechanisms. Specifically, extraversion, agreeableness and neuroticism (Big Five), and psychopathy (Dark Triad) were found to be significant in detecting bullies, achieving up to 96% (precision) and 95% (recall). The emergence of significant personality traits in an experimental study supports existing empirical studies that show the relationships between personality traits and cyberbullying. Highlights: User personalities were used to automatically detect cyberbullying. User personalities based on Big Five and Dark Triad models Random Forest classified users based on Twitter data. Personalities improve detection compared to using only Twitter features. Extraversion, agreeableness, neuroticism and psychopathy had significant roles.