Researchers propose a single hierarchical system for product IDs that can be used in both discovery and search. This system, called Semantic ID, represents products at multiple granularities and is learned from product-content embeddings. It's shown to improve personalized ranking and query reformulation in e-commerce catalogs by aggregating consumer affinity and product performance over the hierarchy. --- Why it matters: This work matters because it shows how a shared semantic product hierarchy can support both recommendation and search, which could lead to more efficient and effective e-commerce systems. This has implications for engineers building recommendation and search systems that need to handle large product catalogs with complex relationships between products. Source: https://arxiv.org/abs/2608.20640