New RL Approach Targets Label Bias in E-Commerce Fraud Detection

An arXiv paper proposes a dual-stream LLM-guided reward design for reinforcement learning that accounts for reject inference, addressing how blocked transactions never receive fraud labels in production fraud models.

Read the full story at arXiv cs.LG ↗

Timeline · 1 report

1. 2026-11-09 05:00 · arXiv cs.LG
   Dual-Stream LLM-Guided Reward Design for Reject-Inference-Aware Reinforcement Learning in E-Commerce Payment Fraud Detection