Ikibazo gihuza ibibazo · Hagati Urwego

Ibicuruzwa bya AI Ibiranga Gukuramo Ikibazo

Kuramo kataloge imirima uhereye kumyandiko n'amashusho mugihe uzigama igishushanyo, ibice, ibimenyetso byatanzwe hamwe nigenzura ryabantu kubintu bitazwi.

Inzira zijyanye no kuyoboraGukuramo ibicuruzwa bya AI
Ikibazo 1 Bya 8

What should a catalog team define before extracting product attributes?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. What should a catalog team define before extracting product attributes?
    • A category-specific schema with types, units and allowed values
    • A prompt asking for any details the model can imagine
    • One universal field list for every industry
    • A rule to publish all high-confidence outputs
  2. What can a product image not reliably establish by itself?
    • That a visible zipper appears in the photograph
    • An item’s hidden fabric composition when no source states it
    • That the image contains a blue object
    • Whether the package shows a printed label
  3. Why retain the source value when normalizing an attribute?
    • To make duplicate products appear unique
    • To avoid validating the normalized field
    • To trace conversions and review how the value was represented originally
    • To remove supplier information
  4. What should a pipeline do when two supplier documents disagree about material?
    • Choose whichever value has more words
    • Average the two material names
    • Publish both values without context
    • Flag the conflict for review instead of silently overwriting
  5. Why treat size or weight as typed fields with units?
    • A value without its unit can be misread or converted incorrectly
    • Units are only needed for images
    • The model can infer the retailer’s preferred unit every time
    • Typed fields prevent supplier errors automatically
  6. What does a confidence score represent?
    • Proof the attribute is factually correct
    • A model estimate that should be checked against evidence
    • An official product certification
    • The supplier’s approval
  7. Which validation is useful for extracted apparel size?
    • Check whether the photo is square
    • Ensure the brand name is capitalized
    • Map the value to the retailer’s allowed size vocabulary and variant
    • Replace every regional size with a single value
  8. How should teams evaluate an attribute extractor?
    • Use one aggregate score for the entire catalog
    • Check only the highest-confidence examples
    • Count how many fields the model outputs
    • Measure accuracy and corrections by field, category and supplier