Analyzing Cyberaggression: Comparative Model Performance on Social Media Comments with Italian Dataset
Vincenzo Gattulli, Donato Impedovo, Alessia Monaco, Lucia Sarcinella · 2024
Cyberbullying is an intentional aggressive act carried out repeatedly through electronic means against a victim unable to defend themself. This phenomenon combines aggressive behavior, repetitive actions, and the victim's inability to defend themself with the permanent nature of online content and easy and wide dissemination. This work focuses on cyberaggression, defined as aggressive online behavior that is distinct from cyberbullying because it is not repeated. A new dataset, “Aggressive Italian Dataset,” and a text feature extraction method were developed and compared with various Shallow Learning and BERT models in Italian — the data collected from Twitter (X) and Instagram concern offenses to famous people. The goal is to expand the dataset to include comments from adult users, simulating the fundamental dynamics of online assaults. BERT appears to have lower Precision, Recall, F1-Score, and Accuracy performance than other Shallow models with our innovative Feature Engineering.