AI for Content Moderation

This chapter contributes to cybersecurity by addressing a subtle form of online harassment that often evades conventional moderation systems, thereby enhancing the robustness of online safety technologies. With the proliferation of social networking services (SNS), the harm caused by abusive comments—including trolling and slander—has become a serious issue, creating a growing demand for automated detection. Among these, comments that use indirect expressions such as sarcasm are particularly difficult to identify. This research focuses on the challenging case of comments containing positive words used sarcastically. We developed a model that integrates fine-tuning of a pre-trained BERT model with additional techniques, such as pattern matching. Experimental results demonstrate that our proposed model outperforms the baseline across four evaluation metrics—Accuracy, Precision, Recall, and F1-score—highlighting its potential to foster safer and more trustworthy online communication environments.