This research advances cyberbullying detection by developing specialized models that distinguish between harassment and defamation, particularly in celebrity-targeted content.

• Creates a novel celebrity cyberbullying dataset covering multiple cyberbullying types
• Uses emotion-adaptive training to improve detection accuracy
• Moves beyond simple binary classification to identify specific forms of harmful content
• Demonstrates how NLP techniques can address complex security threats in social media

This work strengthens online safety by providing more nuanced tools for content moderation, helping platforms better protect users from targeted harassment and reputation damage.

Detecting harassment and defamation in cyberbullying with emotion-adaptive training