Faster Learned Sparse Retrieval with Block-Max Pruning

TL;DR

This paper introduces Block-Max Pruning, a dynamic pruning method for learned sparse retrieval indexes, significantly enhancing efficiency and accuracy in safe and approximate retrieval scenarios.

Contribution

The paper presents a novel block-max pruning strategy specifically designed for learned sparse retrieval indexes, addressing structural differences from traditional models.

Findings

BMP outperforms existing pruning strategies in efficiency

BMP improves tradeoffs between precision and efficiency

Experimental results demonstrate significant retrieval speedups

Abstract

Learned sparse retrieval systems aim to combine the effectiveness of contextualized language models with the scalability of conventional data structures such as inverted indexes. Nevertheless, the indexes generated by these systems exhibit significant deviations from the ones that use traditional retrieval models, leading to a discrepancy in the performance of existing query optimizations that were specifically developed for traditional structures. These disparities arise from structural variations in query and document statistics, including sub-word tokenization, leading to longer queries, smaller vocabularies, and different score distributions within posting lists. This paper introduces Block-Max Pruning (BMP), an innovative dynamic pruning strategy tailored for indexes arising in learned sparse retrieval environments. BMP employs a block filtering mechanism to divide the document space into small, consecutive document ranges, which are then aggregated and sorted on the fly, and fully processed only as necessary, guided by a defined safe early termination criterion or based on approximate retrieval requirements. Through rigorous experimentation, we show that BMP substantially outperforms existing dynamic pruning strategies, offering unparalleled efficiency in safe retrieval contexts and improved tradeoffs between precision and efficiency in approximate retrieval tasks.