AuDisAgent: Training-Free Multimodal Controversy Detection Framework

Researchers propose AuDisAgent, a training-free multi-agent framework for multimodal controversy detection (MCD) that reformulates the task as a dynamic propagation process rather than static representation learning. The framework models audience dissemination through three specialized screening agents: Video Agent, Comment Agent, and Interaction Agent, which conduct initial assessments from visual, textual, and cross-modal perspectives. This approach addresses limitations of prior methods that fail to capture diverse audience perspectives. The framework is designed to support risk management for social video platforms by identifying controversial content in videos and their associated user comments.