Improving cyberbullying detection using Twitter users' psychological features and machine learning

Empirical evidences linking users psychological features such as personality traits and cybercrimes such as cyberbullying are many. This study deals with automatic cyberbullying detection mechanism tapping into Twitter users' psychological features including personalities, sentiment and emotion. User personalities were determined using Big Five and Dark Triad models, whereas machine learning classifiers namely, Naive Bayes, Random Forest and J48 were used to classify the tweets into one of four categories: bully, aggressor, spammer and normal. The Twitter dataset contained 5453 tweets gathered using the hashtag #Gamergate, and manually annotated by human experts. Selected Twitter-based features namely text, user and network-based features were used as the baseline algorithm. Results show that cyberbullying detection improved when personalities and sentiments were used, however, a similar effect was not observed for emotion. A further analysis on the personalities revealed extraversion, agreeableness, neuroticism and psychopathy to have greater impacts in detecting online bullying compared to other traits. Key features were identified using the dimension reduction technique, and integrated into a single model, which produced the best detection accuracy. The paper describes suggestions and recommendations as to how the findings can be applied to mitigate cyberbullying. (C) 2020 Elsevier Ltd. All rights reserved.