A Survey on Multimodal and Sociobehavioral Approaches for Robust Deepfake and AIGenerated Fraud Detection in Social Networks

The rise of generative AI technology has resulted in an increase in deepfakes and AI-generated
fraud on social networks, posing considerable risks to trust, security, and information integrity.
Conventional unimodal methods that concentrate on visual, auditory, or textual signals
frequently falter in the face of advanced forgeries. To tackle this issue, researchers have created
multimodal approaches that combine content-based signals with sociobehavioral tendencies,
providing enhanced identification capabilities. This survey presents a classification of
detection strategies—unimodal, multimodal, sociobehavioral, and hybrid approaches—
accompanied by a critical evaluation of recent developments. We emphasize datasets,
assessment measures, obstacles including adversarial robustness and scalability, and propose
future research paths, underscoring the necessity of integrating multimodal and
sociobehavioral techniques for comprehensive fraud detection in social networks