Article

Emotion-Aware Detection of Cyberbullying in Visual Social Media Content

---

Abstract

Cyberbullying remains a persistent and evolving threat in digital society, often manifesting in subtle, emotionally charged, and context-dependent forms that evade traditional detection mechanisms. This study explores the effectiveness of an emotion-aware approach to cyberbullying detection by analyzing a multimodal dataset of 5,793 social media posts, each annotated with labels for emotion, sentiment polarity, sarcasm, harmfulness, and target type. The findings reveal that negative sentiment dominates the dataset (2,499 posts), with emotionally intense categories such as Disgust (913 instances), Ridicule (687), and Anger (653) strongly associated with bullying content. Notably, 3,188 posts (55.0%) were labeled as Bully, and 3,072 posts were found to target specific Individuals, confirming the personal nature of digital aggression. Sarcasm was present in 1,179 posts (20.3%), and these were disproportionately represented in the Partially-Harmful class (2,338 posts), suggesting that covert hostility is a prevalent form of abuse in online discourse. The analysis demonstrates that nearly 49.6% of content carries some degree of harmful potential, either explicitly or implicitly, reinforcing the limitations of binary classifiers. These findings underscore the need for fine-grained, affect-sensitive models capable of capturing emotional and rhetorical complexity in social media content. The study provides a foundational empirical basis for the development of multimodal, emotion-aware cyberbullying detection systems that are more attuned to the nuanced realities of online harm.

Keywords: Emotion-aware detection, cyberbullying, sentiment analysis, sarcasm, multimodal social media

How to Cite:

Rahma, F. & Ulfah, S., (2025) “Emotion-Aware Detection of Cyberbullying in Visual Social Media Content”, Journal of Digital Society 1(4), 299-313. doi: https://doi.org/10.63913/jds.v1i4.44