Context
Necula and Sireteanu compared lexical, dense, hybrid and cross-encoder retrieval methods on WANDS and a held-out US ESCI subset. Their peer-reviewed study reports significant nDCG@10 improvements for hybrid and cross-encoder methods over BM25 on both benchmarks, while pure dense retrieval had no significant overall gain after Holm correction.
What changed
The September 29, 2026 Scientific Reports study supplies peer-reviewed controlled comparisons of hybrid lexical-semantic product retrieval, pure dense retrieval and cross-encoder reranking across two e-commerce benchmarks.
Why it matters
AI shopping discovery must satisfy explicit attributes as well as semantic intent. Retrieval pipelines should be evaluated for query-type relevance and not assume dense semantic matching is universally superior to keyword matching.
Visibility OS interpretation
Visibility OS interpretation: Hybrid lexical and semantic matching can be a practical evaluation baseline for product retrieval. The paper is an offline e-commerce search study, not evidence of consumer ChatGPT, Bing, Gemini or Perplexity product ranking or citation behavior.
What remains uncertain
How do these results translate to live AI shopping agents, changing product catalogs, purchase satisfaction and multilingual attribute-heavy queries? How robust are the findings under different retrieval budgets?