Evaluating Cyberbullying Detection Algorithm Performance in Text and Image Analysis

Cyberbullying has become a widespread concern in our society, and its detection has grown in significance. In recent years, machine learning algorithms have exhibited considerable potential in detecting cyberbullying. This paper presents an approach for detecting it employing Natural Language Processing and image analysis methods to analyze both textual and visual content. The study utilized an online comment dataset from Twitter and a self-created image dataset, where labeling was done for cyberbullying and non-cyberbullying, while tweets underwent detailed labeling. Multiple models were tested to determine the most suitable one. The Support Vector Machine model achieved an accuracy rate of 81.4% on our text dataset, according to our experimental results. The model achieved a remarkable performance with Area Under the Curve scores exceeding 0.97 in classifying comments based on age, ethnicity, gender, and religion. For cyberbullying detection in image data, pre-trained models with CNN architecture were utilized. Regrettably, our experiments in this regard failed to produce satisfactory out- comes, possibly due to varying interpretations of visual content among individuals. We will further discuss this issue in our paper. Overall, our approach offers a potentially effective solution for identifying instances of cyberbullying in textual content, which could be utilized in the development of tools and systems aimed at preventing cyberbullying and promoting online safety.