Analysis and detection of labeled cyberbullying instances in Vine, a video-based social network

Abstract

The last decade has experienced an exponential growth of popularity in online social networks. This growth in popularity has also paved the way for the threat of cyberbullying to grow to an extent that was never seen before. Online social network users are now constantly under the threat of cyberbullying from predators and stalkers. In our research paper, we perform a thorough investigation of cyberbullying instances in Vine, a video-based online social network. We collect a set of media sessions (shared videos with their associated meta-data) and then label those using CrowdFlower, a crowd-sourced website for cyberaggression and cyberbullying. We also perform a second survey that labels the videos’ contents and emotions exhibited. After the labeling of the media sessions, we provide a detailed analysis of the media sessions to investigate the cyberbullying and cyberaggression behavior in Vine. After the analysis, we train different classifiers based upon the labeled media sessions. We then investigate, evaluate and compare the classifers’ performances to detect instances of cyberbullying.