Proof-of-Concept Experiments for the Fine-Grained Classification of Cyberbullying Events
Cynthia Van Hee, Ben Verhoeven, Els Lefever, Guy De Pauw, Véronique Hoste, Walter M. P. Daelemans · Ghent University Academic Bibliography (Ghent University) · 2015
In the current era of online interactions, both positive and negative experiences are abundant on the web. As in real life, these negative experiences can have a serious impact on youngsters. Recent research reports cybervictimization rates among teenagers between 3% and 24% (Olweus, 2012; Hinduja & Patchin, 2012). In the research project AMiCA, we strive to automatically detect harmful content such as cyberbullying on social networks. We collected data from social networking sites and by simulating cyberbullying events with volunteer youngsters. This dataset was annotated for a number of fine-grained categories related to cyberbullying such as insults and threats. More broadly, the severity of cyberbullying, as well as the author's role (harasser, victim or bystander) were defined. We present the results of our preliminary experiments where we try to determine whether an online message is part of a cyberbullying event. Moreover, we explore the feasibility to classify online posts in five more fine-grained categories. For the binary classification task of cyberbullying detection, we obtain an F-score of 54.71%. F-scores for the more fine-grained classification tasks vary between 20.51% and 55.07%.