Multi-Stage Document Ranking with BERT

TL;DR

This paper introduces multi-stage document ranking models using BERT, achieving a balance between search quality and latency, and demonstrating competitive results on large-scale datasets.

Contribution

It proposes two BERT-based ranking variants, monoBERT and duoBERT, integrated into a multi-stage architecture for efficient and effective document retrieval.

Findings

Achieves state-of-the-art or comparable results on MS MARCO and TREC CAR datasets.

Demonstrates effective trade-offs between ranking quality and latency.

Ablation studies highlight the importance of each component in the system.

Abstract

The advent of deep neural networks pre-trained via language modeling tasks has spurred a number of successful applications in natural language processing. This work explores one such popular model, BERT, in the context of document ranking. We propose two variants, called monoBERT and duoBERT, that formulate the ranking problem as pointwise and pairwise classification, respectively. These two models are arranged in a multi-stage ranking architecture to form an end-to-end search system. One major advantage of this design is the ability to trade off quality against latency by controlling the admission of candidates into each pipeline stage, and by doing so, we are able to find operating points that offer a good balance between these two competing metrics. On two large-scale datasets, MS MARCO and TREC CAR, experiments show that our model produces results that are either at or comparable to the state of the art. Ablation studies show the contributions of each component and characterize the latency/quality tradeoff space.