Analysis and Detection of Cyberbullying Targeting Minors on Social Media Using Behavior Logs

Abstract

With the spread of smartphones and social media, cyberbullying, a new kind of bullying perpetrated over electronic media such as email and social media, has increased. In this study, we address two analytical challenges by using social media usage logs of minors in the real world: (1) Is there a significant behavioral trend in cyberbullying? (2) How useful are non-text activity logs for detecting bullying victims? Our first analysis yielded interesting findings about cyberbullying behavior such as that cyberbullying is likely to take place at night, and that victims are likely to use the app until midnight. Furthermore, the results of the second analysis indicated that we succeeded in developing a meaningful model with a Macro-F1 value of 0.760. We showed that the features obtained from the behavior data are useful for victim detection.