Researchers have developed MM-FGDNet, a novel multimodal framework designed to detect coordinated abnormal behaviors in live-streaming platforms. This system integrates video, text, audio, and user behavior data, mapping them into a unified temporal semantic space. It specifically targets the progression of suspicious activities and identifies coordinated interactions among user groups and automated accounts, aiming for early detection and reduced false alarms. Experiments on real-world datasets demonstrated MM-FGDNet's superior performance compared to existing methods, achieving high metrics including an AUC of 0.927 and an Early Detection Score of 0.689. AI
IMPACT Provides a new method for identifying sophisticated fraudulent activities in live-streaming environments, potentially improving platform integrity and user trust.
RANK_REASON The item is a research paper published on arXiv detailing a new technical method. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
