Active Learners as Efficient PRP Rerankers

TL;DR

This paper introduces an active learning framework for pairwise ranking with noisy judgments, improving ranking quality under call constraints by reducing bias and cost.

Contribution

It reframes PRP reranking as an active learning problem, proposing a noise-robust method with a randomized-direction oracle that enhances ranking accuracy.

Findings

Active rankers improve NDCG@10 in call-constrained settings.

The randomized-direction oracle reduces systematic bias to zero-mean noise.

The approach requires only a single LLM call per pair, lowering costs.

Abstract

Pairwise Ranking Prompting (PRP) elicits pairwise preference judgments from an LLM, which are then aggregated into a ranking, usually via classical sorting algorithms. However, judgments are noisy, order-sensitive, and sometimes intransitive, so sorting assumptions do not match the setting. Because sorting aims to recover a full permutation, truncating it to meet a call budget does not produce a dependable top-K. We thus reframe PRP reranking as active learning from noisy pairwise comparisons and show that active rankers are drop-in replacements that improve NDCG@10 per call in the call-constrained regime. Our noise-robust framework also introduces a randomized-direction oracle that uses a single LLM call per pair. This approach converts systematic position bias into zero-mean noise, enabling unbiased aggregate ranking without the cost of bidirectional calls.