Reproduce “One Hierarchy, Two Systems: Semantic Product IDs for Discovery-Surface Ranking and Search-Page Query Reformulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Steven Xu, Sanjyot Thete, Saathvik Dirisala, Raghav Saboo, Nimesh Sinha, Leo Shao, Elyse Winer, Sudeep Das and 2 more
DoorDash researchers developed a learned hierarchical Semantic ID (SID) representation from product content, applying it to enhance personalized item ranking on discovery surfaces and improve query reformulation on search pages. This system led to increased subtotal and add-to-cart engagement in ranking, and higher purchase MRR and reduced search effort in query reformulation.
Key insights
- A single learned hierarchical SID representation can serve as a robust, transferable semantic prior across distinct e-commerce discovery tasks, effectively bridging fragmented identifiers and coarse taxonomies.
- SIDs enable the aggregation of behavioral evidence over semantic product concepts, leading to more effective cross-merchant preference transfer and finer-grained personalization in ranking.
- The hierarchical structure of SIDs supports both fine-grained conceptual refinements and lateral pivots in search, providing a more accurate and assortment-aware mechanism for understanding user intent and generating suggestions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.