compere explores how to achieve high-quality rankings with minimal human feedback. It implements multi-armed bandit algorithms optimised for pairwise comparison settings, studying applications in search ranking, recommendation systems, and tournament design.
Rank items with fewer pairwise comparisons — search evaluation, tournament design, recommendation feedback.
compere is one option in a category that includes Bradley-Terry, TrueSkill, Elo, Plackett-Luce , and preference-learning libraries. Our Compare page has the full side-by-side.