Identification of Sexual Harassment in Social Media Comments Using IndoBERT and Support Vector Machine

The rise of social media platforms has facilitated user interaction but has also led to increased incidents of online harassment, particularly sexual harassment. This form of harassment often manifests through sexualised and demeaning comments, disproportionately targeting women. Identifying such comments is challenging due to the high volume of content and the subjective nature of harassment. Manual identification is time-consuming and inconsistent. Therefore, an automated system to identify sexually harassing comments is needed for more effective management of social media content. This research introduces a system combining the IndoBERT embedding method and the Support Vector Machine (SVM) algorithm to identify sexually harassing comments in Indonesian social media content. The dataset comprises 3500 comments from Instagram and YouTube. Evaluation using a Confusion Matrix reveals an accuracy rate of 91.1%, significantly outperforming traditional keyword-based approaches. The system demonstrated strong performance in identifying implicit sexual harassment, which is often overlooked by manual moderation systems. These findings suggest that integrating IndoBERT and SVM provides a practical and scalable solution for real-time content moderation, with potential applications in enhancing the safety of online spaces.