Knowledge base
Search evaluation: from individual examples to reproducible quality.
Good search is not judged on one beautiful demo query, but on a fixed representative test set and clear success criteria.
Build a query set
Include top queries, long-tail, brands, SKUs, typos, categories and problem queries.
Capture relevance
Use relevance judgments to determine which results are strong, partially relevant or unsuitable.
Test regressions
Protect critical exact matches when ranking, product data, or models change.
Combine offline and online
Use CTR, reformulations and conversion in addition to offline evaluation, as commercial signals are also affected by price and stock.
Immerse yourself in Search Evaluation →