Robust Active Ranking from Sparse Noisy Comparisons.
Year: 2015
Type: preprint
Abstract: From sporting events to sociological surveys, ranking from pairwise comparisons is a tool of choice for many applications. When certain pairs of items are difficult to compare, outcomes can be noisy, and it is necessary to develop robust strategies. In this work, we show how a simple active sampling scheme that uses a standard black box sorting algorithm enables the efficient recovery of the ranking, achieving low error with sparse samples. Both in theory and practice, this active strategy performs systematically better than selecting comparisons at random. As a detour, we show a link between Rank Centrality, a recently proposed algorithm for rank aggregation, and the ML estimator for the Bradley-Terry model. This enables us to develop a new, provably convergent iterative algorithm for computing the ML estimate.
Source: arXiv (Cornell University)
Authors: Lucas Maystre, Matthias Grossglauser
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W2152408869