Leveraging IndoBert for CyberBullying Classification on Social Media
Kelvin Chandra, Kristo Amadeus Prasetya, Rizky Dwi Saputra, Muhammad Fikri Hasani · 2024
Cyberbullying is becoming an increasingly concerning issue, with a significant increase in cases every day. These incidents have profound negative impacts on victims, including emotional stress, mental health problems, and social withdrawal. This research aims to comprehensively examine the development and prevalence of cyberbullying cases in Indonesia. By utilizing sentiment analysis on Instagram comments, this study evaluates the accuracy and effectiveness of two prominent algorithms: Neural Network and IndoBERT (Indonesian Bidirectional Encoder Representations from Transformers). The research findings indicate that the IndoBERT algorithm demonstrates superior performance, achieving high accuracy in identifying instances of cyberbullying. Specifically, IndoBERT shows an accuracy of $\mathbf{9 2 \%}$, precision of $\mathbf{9 2 \%}$, recall of $\mathbf{9 2 \%}$, and an F1 score of $\mathbf{9 2 \%}$. These results underscore the potential of advanced machine learning models in enhancing the detection of harmful online behavior. The study also recommends that future research should utilize larger datasets to further improve the accuracy and reliability of analysis.