CoRT: Complementary Rankings from Transformers

TL;DR

CoRT is a neural first-stage ranking model that enhances traditional term-based methods like BM25 by leveraging pretrained language models, significantly improving candidate recall and enabling efficient passage retrieval.

Contribution

This paper introduces CoRT, a neural ranking model that complements BM25 using contextual embeddings, improving recall without adding significant query-time delay.

Findings

CoRT increases candidate recall when combined with BM25.

Subsequent re-rankers perform better with fewer candidates.

Passage retrieval with CoRT achieves low latency.

Abstract

Many recent approaches towards neural information retrieval mitigate their computational costs by using a multi-stage ranking pipeline. In the first stage, a number of potentially relevant candidates are retrieved using an efficient retrieval model such as BM25. Although BM25 has proven decent performance as a first-stage ranker, it tends to miss relevant passages. In this context we propose CoRT, a simple neural first-stage ranking model that leverages contextual representations from pretrained language models such as BERT to complement term-based ranking functions while causing no significant delay at query time. Using the MS MARCO dataset, we show that CoRT significantly increases the candidate recall by complementing BM25 with missing candidates. Consequently, we find subsequent re-rankers achieve superior results with less candidates. We further demonstrate that passage retrieval using CoRT can be realized with surprisingly low latencies.