The paper announces CareGuard, an early-warning framework for detecting cyberbullying-related content. The framework utilizes zero-shot semantic labeling combined with fine-tuned transformer-based models – BERT, DistilBERT, and RoBERTa – to achieve robust and context-aware classification across cyberbullying categories. An emotion-aware filtering mechanism is incorporated alongside cosine similarity-based semantic screening. This allows the system to prioritize semantically relevant and emotionally salient content. Experimental results on benchmark datasets demonstrate a balance between detection accuracy and computational efficiency. This framework has potential for deployment in healthcare systems, mental health monitoring, and online safety applications.
Source: https://arxiv.org/abs/2609.09735