Deja Vu in Plots: Leveraging Cross-Session Evidence with Retrieval-Augmented LLMs for Live Streaming Risk Assessment
The paper 'Deja Vu in Plots' presents a dual system for live streaming risk assessment. It uses PatchNet to identify local key moments and an LLM to find shared patterns across multiple sessions. This allows the AI to detect coordinated, subtle fraudulent behaviors that appear benign individually but share a common behavioral chain. The system achieves state-of-the-art performance in proactive risk management.