Composite Re-Ranking for Efficient Document Search with BERT

TL;DR

BECR is a novel re-ranking method that combines deep contextual token interactions with traditional lexical features, significantly improving efficiency and relevance in document search without expensive online BERT computations.

Contribution

This paper introduces BECR, a composite re-ranking scheme that separates token encoding from online inference, enabling faster yet effective document ranking with BERT-based models.

Findings

BECR achieves high relevance in ad-hoc ranking tasks.

BECR significantly reduces online inference time.

BECR outperforms other neural ranking models on TREC datasets.

Abstract

Although considerable efforts have been devoted to transformer-based ranking models for document search, the relevance-efficiency tradeoff remains a critical problem for ad-hoc ranking. To overcome this challenge, this paper presents BECR (BERT-based Composite Re-Ranking), a composite re-ranking scheme that combines deep contextual token interactions and traditional lexical term-matching features. In particular, BECR exploits a token encoding mechanism to decompose the query representations into pre-computable uni-grams and skip-n-grams. By applying token encoding on top of a dual-encoder architecture, BECR separates the attentions between a query and a document while capturing the contextual semantics of a query. In contrast to previous approaches, this framework does not perform expensive BERT computations during online inference. Thus, it is significantly faster, yet still able to achieve high competitiveness in ad-hoc ranking relevance. Finally, an in-depth comparison between BECR and other start-of-the-art neural ranking baselines is described using the TREC datasets, thereby further demonstrating the enhanced relevance and efficiency of BECR.