What question did this study set out to answer?

The aim is to explore the use of Large Language Models for detecting and predicting cyberbullying effectively.

March 23, 2026Open Access

Leveraging Large Language Models for Cyberbullying Prediction and Detection: A Multimodal Approach to Proactive Prevention and Policy Support.

Key Points

The aim is to explore the use of Large Language Models for detecting and predicting cyberbullying effectively.
Review challenges in cyberbullying definitions and prevention.
Utilize Google Gemini model for experiments.
Evaluate text and image-based datasets for analysis.
Implement zero shot learning for detection and prediction.
LLM achieved 82% balanced accuracy in cyberbullying detection.
75.8% accuracy in multimodal prediction.
First exploration utilizing LLMs for cyberbullying prevention.

Abstract

Cyberbullying (CB) continues to outpace effective prevention, intervention and aftercare measures posing a growing public health concern as digital systems become central to youth life. Persistent definitional ambiguities hinder evidence-based research and policy development. This study reviews these challenges and explores a technical solution using Large Language Models (LLMSs) for CB detection and prediction. Leveraging Google Gemini model, experiments evaluated text and image-based datasets for sentiments and behavioral analysis. LLM achieved 82% balanced accuracy in detection through zero shot learning and 77.8% in multimodal prediction, marking the first known exploration of this capability. Findings suggest that LLMs can enable flexible, scalable and proactive CB prevention strategies, offering a promising complement to existing policy and educational approaches.

Leveraging Large Language Models for Cyberbullying Prediction and Detection: A Multimodal Approach to Proactive Prevention and Policy Support.

Key Points

Abstract

Cite This Study