Multivector Reranking in the Era of Strong First-Stage Retrievers

TL;DR

This paper improves multivector retrieval efficiency by replacing token-level gather with a learned sparse retriever, integrating inference-free methods, and optimizing reranking, achieving significant speedups without quality loss.

Contribution

It introduces a two-stage retrieval pipeline using a learned sparse retriever for candidate gathering, combined with inference-free methods and candidate pruning, enhancing efficiency while maintaining effectiveness.

Findings

Achieves over 24x speedup compared to state-of-the-art systems.

Maintains retrieval quality with up to 1.8x efficiency improvements.

Replaces token-level gather with a learned sparse retriever for better candidate selection.

Abstract

Learned multivector representations power modern search systems with strong retrieval effectiveness, but their real-world use is limited by the high cost of exhaustive token-level retrieval. Therefore, most systems adopt a \emph{gather-and-refine} strategy, where a lightweight gather phase selects candidates for full scoring. However, this approach requires expensive searches over large token-level indexes and often misses the documents that would rank highest under full similarity. In this paper, we reproduce several state-of-the-art multivector retrieval methods on two publicly available datasets, providing a clear picture of the current multivector retrieval field and observing the inefficiency of token-level gathering. Building on top of that, we show that replacing the token-level gather phase with a single-vector document retriever -- specifically, a learned sparse retriever (LSR) -- produces a smaller and more semantically coherent candidate set. This recasts the gather-and-refine pipeline into the well-established two-stage retrieval architecture. As retrieval latency decreases, query encoding with two neural encoders becomes the dominant computational bottleneck. To mitigate this, we integrate recent inference-free LSR methods, demonstrating that they preserve the retrieval effectiveness of the dual-encoder pipeline while substantially reducing query encoding time. Finally, we investigate multiple reranking configurations that balance efficiency, memory, and effectiveness, and we introduce two optimization techniques that prune low-quality candidates early. Empirical results show that these techniques improve retrieval efficiency by up to 1.8$\times$ with no loss in quality. Overall, our two-stage approach achieves over $24\times$ speedup over the state-of-the-art multivector retrieval systems, while maintaining comparable or superior retrieval quality.