Investigating Text and Image Features for the Detection of Cyberbulling
Julia E. Heine, Kristina Schaaff, Tim Schlippe · 2024
A very serious issue throughout social media platforms and cultural groups on the internet is the phenomenon of cyberbullying, an emerging form of victimization brought about through the digital age [1].In this paper, we investigate traditional and new features to detect cyberbullying in text messages which contain images.Our best text-based cyberbullying detection system achieves an accuracy of 66.7% on a balanced subset of the MMSH150K dataset [2].The system uses a combination of text vector features, semantic features, list lookup features, emoji features, error-based features, and AI feedback features.Our best image-based cyberbullying detection system, which leverages image text features, obtains an accuracy of 52.5%.This shows that text vector features and image vector features, which are mostly used for this task in related work, do not necessarily lead to the best results and our new features contribute a part to improving the performance.