Beyond Artificial Intelligence for Search
summary
Claims that for structured e‑commerce data a hand‑tuned search configuration — using boosts and function_scores, domain taxonomy and synonyms, redirects and business landing pages — outperforms ML ranking, which is costly, brittle and a black box. Uses ML where it fits: crawling/indexing, OCR cleanup, dynamic synonyms and NLP. Proposes a social interaction relevance model as an alternative to PageRank and heavy ML and describes a built collaborative engine that surfaces results like a mechanical turk powered by users and domain experts.
AI-written summary · about