Whole-Pool Setwise Reranking with Long-Context Language Models

2026-06-01

AI Summary

Traditional large language models (LLMs) suffer from limited context lengths, necessitating multiple sequential calls for passage re-ranking—a process that is both inefficient and costly. This work proposes a holistic pool-based re-ranking approach leveraging long-context LLMs, introducing an innovative DualEnd bidirectional identification strategy that simultaneously pinpoints the most and least relevant passages within a single model invocation. By constructing the ranking from both ends in concert, this method pioneers the use of long-context capabilities for end-to-end re-ranking, drastically reducing the number of model calls and computational overhead. Evaluated across two benchmarks with nine open-source models, the approach achieves effective re-ranking of 100 candidate passages using only 50 calls—nearly halving the invocation count compared to per-passage methods—while maintaining high ranking effectiveness and output reliability.

Abstract

Previous LLM-based passage re-rankers are often expensive and slow because the input context constraints require the LLM to make many dependent model calls. We study how recent long-context LLMs change this problem: when the full set of retrieved candidate passages can be shown to the model at once, ranking no longer has to be reconstructed from many overlapping local comparisons. We propose Whole-Pool Setwise re-ranking, where each call considers all currently unranked candidate passages, and introduce DualEnd, which identifies both the most and least relevant passages in one call. By filling the ranking from both ends, DualEnd ranks 100 candidates with 50 serial LLM calls, compared with 99 calls for comparable one-passage-at-a-time whole-pool methods. Experiments with nine open-weight LLMs on two passage re-ranking benchmarks, measuring effectiveness, call count, token use, runtime, and output reliability shows that long context is not merely more prompt space, but an opportunity to make LLM re-rankers both effective and efficient.

Problem

• passage re-ranking
• long-context LLMs
• whole-pool ranking
• efficient inference
• setwise comparison

Innovation

• Whole-Pool Setwise Reranking
• Long-Context LLMs
• DualEnd
• Passage Re-ranking
• Efficient Inference