When might top-k pair construction be preferable to uniform mean pair sampling in ranking?
Classic MLLevel: IntermediateXGBoost · Learning to rank
When might top-k pair construction be preferable to uniform mean pair sampling in ranking?
Hints
- Which positions matter to the product?
- Coverage and head emphasis are different objectives.
Answer
When the product mainly evaluates a short top-ranked list, emphasizing its head can align learning with the target. Mean sampling covers more of the list; compare on the actual cutoff and enough queries.
Source and licence: XGBoost · Learning to rank · Apache-2.0 · answer: Quantoros editorial