Direct Learning to Rank and Rerank

TL;DR

This paper critiques current learning-to-rank algorithms for relying on convex proxies, proposes exact reranking methods via mathematical programming, and empirically analyzes their effectiveness.

Contribution

It introduces a novel approach to learning-to-rank using exact reranking with mathematical programming, addressing limitations of convex proxy methods.

Findings

Convex proxies lead to poor approximations in ranking tasks.

Relaxed exact reranking problem shares the same optimal solution.

Empirical results demonstrate the effectiveness of the proposed reranking approach.

Abstract

Learning-to-rank techniques have proven to be extremely useful for prioritization problems, where we rank items in order of their estimated probabilities, and dedicate our limited resources to the top-ranked items. This work exposes a serious problem with the state of learning-to-rank algorithms, which is that they are based on convex proxies that lead to poor approximations. We then discuss the possibility of "exact" reranking algorithms based on mathematical programming. We prove that a relaxed version of the "exact" problem has the same optimal solution, and provide an empirical analysis.