domains / families / research / recsys-experiments

recsys-experiments

separate

Use when designing or auditing ACM RecSys experiments centered on offline-versus-online evaluation — temporal splits, equal-budget baseline tuning, full-ranking versus sampled metrics, off-policy estimators (IPS, SNIPS, doubly robust), A/B tests, exposure and popularity bias, seeds and variance, and matching each recommendation claim to its evidence.

When to use

• designing RecSys experiments
• auditing RecSys experiments
• offline-versus-online evaluation
• temporal splits
• equal-budget baseline tuning
• full-ranking versus sampled metrics
• off-policy estimators
• matching recommendation claims to evidence

Skill page

Source