Using GPT-4o-mini as an Entity Resolution Judge: 95% Precision for $0.04

Entity resolution - finding records that refer to the same real-world entity - is one of the oldest problems in data management. And for structured data (names, addresses, numbers), we’ve largely solved it. Modern fuzzy matching algorithms routinely achieve 95%+ accuracy. Product data is a different story. THE PRODUCT MATCHING PROBLEM Consider matching these two records: Source A: “Sony Cyber-shot DSC-T77 Silver” Source B: “Sony - Cyber-shot 10.1-Megapixel Digital Camera - Silver” A human instantly recognizes these as the same product.