Guide-linked quiz · Medium Level

AI Site Search for Online Stores Quiz

Test AI store search: query understanding, retrieval and reranking, synonyms, hard filters, catalog facts, relevance metrics, click bias, and privacy.

Question 1 of 8

Why can a two-stage search system use both retrieval and reranking?

Preview every question8 questions

Answers and explanations appear as you take the quiz.

  1. Why can a two-stage search system use both retrieval and reranking?
    • Reranking creates the catalog from scratch
    • Retrieval finds candidates; a reranker orders those candidates using additional relevance signals
    • Retrieval is used only after the final ranking
    • Both stages guarantee exact product identity
  2. A synonym rule maps “sneakers” to “running shoes.” What risk should be tested?
    • The rule will delete all shoe products
    • A synonym changes the product’s material
    • A synonym prevents spelling correction
    • The expansion may return products that are not relevant to some sneaker queries
  3. A product appears relevant but does not meet the selected size filter. What should the search system do?
    • Exclude it because explicit filters are hard constraints
    • Promote it if its click rate is high
    • Ignore the filter when an AI model is confident
    • Show it as a synonym
  4. Why can click logs create a feedback loop in product search?
    • Clicks are collected before products are shown
    • A click proves a product is available
    • Top-ranked products receive more exposure and clicks, which can reinforce their future rank
    • Click data remove position bias automatically
  5. Which offline metric rewards relevant products appearing near the top of results?
    • Pixel accuracy
    • NDCG
    • BLEU
    • Word error rate
  6. Why test long-tail and zero-result queries?
    • Average performance can hide failures on less common needs and missing catalog coverage
    • Long-tail queries always have more clicks
    • A zero-result query proves a user is not interested
    • Common queries are never relevant
  7. A model infers that a jacket is waterproof from marketing copy, but the catalog has no verified waterproof attribute. What is the risk?
    • The model will improve inventory accuracy
    • The jacket’s stock count changes
    • A filter cannot display products
    • Search may make an unsupported product claim based on ambiguous text
  8. Which measure complements conversion when evaluating search?
    • Model training time only
    • Number of vector dimensions
    • Zero-result searches, reformulations, returns, and complaints
    • Page color preference

Keep testing yourself

More quizzes picked for your level.

View all quizzes