Paper introduces post-stratification for heavy-tailed ranking metrics

Arxiv CS.LG (machine learning)

The paper proposes post-stratification to reduce variance in heavy-tailed monetization metrics such as app revenue or creator earnings. It addresses the issue where a small fraction of users dominate both mean and variance, improving the reliability of online evaluation of ranking systems.

Arxiv CS.LG (machine learning)·Neeti PokharnaOlivier JeunenYatharth SarafAleksei Ustimenko

7 days ago