TREC 2025 Product Search Track releases new e-commerce retrieval benchmark
First dataset that separates complementary from related products for smarter recommendations
TREC (the Text Retrieval Conference) has published its 2025 Product Search and Recommendation Track overview, authored by Dean Alvarez, Surya Kallumadi, Daniel Campos, and four other researchers. Building on the 2023 and 2024 iterations, the track addresses a long-standing gap: despite e-commerce's heavy reliance on search and recommendation, no high-quality dataset existed for evaluating end-to-end retrieval quality. The 2025 edition introduces a revised, expanded benchmark designed to reflect how consumers actually browse and compare products across large, diverse catalogs.
The track comprises two tasks: query expansion, which tests a system's ability to enrich a user's search query with relevant context, and related-product recommendation, where the dataset innovatively annotates product relationships to explicitly distinguish complementary items (e.g., a phone case for a phone) from related alternatives (e.g., a different phone). This nuance is crucial: understanding why two products are linked—rather than just that they are—enables more intelligent, context-aware recommendations. The organizers anticipate the data will power better search and recommendation applications, serving as a foundation for conversational product discovery experiences.
- TREC 2025 track continues from 2023/2024, adding a revised benchmark for end-to-end e-commerce retrieval evaluation
- Two tasks: query expansion and related-product recommendation, with annotated data distinguishing complementary vs related products
- Aimed at enabling more accurate search and conversational product discovery experiences, filling a missing benchmark gap
Why It Matters
E-commerce teams can use this dataset to build search and recommendation systems that genuinely understand purchase intent and product relationships.