Question-to-Knowledge (Q2K): Multi-Agent Generation of Inspectable Facts for Product Mapping

The introduction of the Question-to-Knowledge (Q2K) framework marks a significant advancement in ecommerce, particularly in resolving the persistent challenge of identifying whether two product listings correspond to the same Stock Keeping Unit (SKU). Traditional methods often fail due to variations in product names and lack of explicit identifiers, leading to misclassifications. Q2K employs a multi-agent system comprising a Reasoning Agent that formulates disambiguation questions, a Knowledge Agent that conducts focused web searches to answer these questions, and a Deduplication Agent that minimizes redundancy by reusing validated reasoning. This innovative approach not only enhances accuracy but also ensures consistency in SKU mapping. Experiments conducted on real-world consumer goods datasets reveal that Q2K surpasses strong baselines, achieving higher accuracy and robustness, particularly in challenging scenarios such as bundle identification and brand origin disambiguation. This …