EPM-RL: Reinforcement Learning for On-Premise Product Mapping in E-Commerce

Abstract

Product mapping decides whether two e-commerce listings refer to the same product. We propose EPM-RL, an on-premise framework that distills LLM comparison reasoning into a local model via PEFT and RL, improving balanced matching under privacy and cost constraints. On an internal benchmark, EPM-RL yields the strongest F1 by boosting recall while maintaining competitive precision.