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

EPM-RL is a reinforcement-learning framework for building accurate and efficient on-premise e-commerce product mapping models. Product mapping determines whether two listings refer to the same product, crucial for price monitoring and channel visibility. Sellers often use promotional keywords, platform-specific tags, and bundle descriptions, causing the same product to appear under many names. Existing LLM-based and multi-agent frameworks improve robustness but rely on expensive external APIs and complex orchestration, hindering large-scale deployment in privacy-sensitive enterprise settings. EPM-RL distills high-cost agentic reasoning into a trainable in-house model, starting from a curated dataset. The framework addresses cost and privacy concerns by enabling on-premise deployment without external API calls.