Date Approved
9-24-2025
Graduate Degree Type
Thesis
Degree Name
Applied Computer Science (M.S.)
Degree Program
School of Computing and Information Systems
First Advisor
Sara Sutton
Second Advisor
Rahat Ibn Rafiq
Third Advisor
Rajvardhan Patil
Academic Year
2024/2025
Abstract
Cyberbullying poses a significant challenge on social media, where traditional detection systems struggle with the nuanced and dynamic nature of online abuse. This thesis proposes an integrated framework that combines large language model (LLM)-based detection with generative decoy responses to enable real-time protection for victims on platforms like Instagram and WhatsApp. Using models such as Mistral 7B, GPT-3.5 Turbo, and Phi-3 Mini, prompt-based one-shot learning achieved 87% detection accuracy and a 5% F1-score improvement over zero-shot approaches, demonstrating robust identification of text-based bullying. A novel synthetic dataset of 98 multi-turn conversations, designed with diverse subtypes and evaluated for realism, addressed the scarcity of context-rich training data. The decoy response system generated safe, context-aware replies. Ethical considerations, including bias mitigation and user consent, were prioritized. Future work includes multilingual expansion and real-time deployment, advancing safer digital communities.
ScholarWorks Citation
Islam, Mohammad Shafiqul, "Cyberbullying Defensive Strategy in Social Media Sessions via Machine Learning and Cyber Deception" (2025). Masters Theses. 1168.
https://scholarworks.gvsu.edu/theses/1168
