SUPPORT VECTOR MACHINE IMPLEMENTATION FOR RACIAL CYBERBULLYING SENTIMENT ANALYSIS IN YOUTUBE COMMENTS

• Main Author: Made Alit Adinugraha, Dewa
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86873
• Institution: Institut Teknologi Bandung

This study focuses on developing a model for detecting and classifying cyberbullying, particularly related to hate speech based on ethnicity, religion, race, and intergroup relations, using Support Vector Machine (SVM) and TF-IDF weighting. Data was collected from YouTube comments and categorized into two classes: non-cyberbullying and Racial-based cyberbullying. The model was designed to classify negative sentiment comments containing cyberbullying. A confusion matrix evaluation was applied to assess model accuracy, precision, and recall, using k-fold cross-validation. Results showed varying model performance across each fold, with an overall average accuracy of 85%, F1-score of 89%, and recall of 90%, with the best performance observed in 3 rd folds and 4th. The model struggled to detect Racial-based cyberbullying in fold 5th, affecting its overall performance. This study demonstrates that SVM is fairly effective in detecting cyberbullying, though improvements are needed to enhance classification accuracy and ensure that various forms and categories of cyberbullying are adequately identified.