Efficient Document Ranking with Learnable Late Interactions

TL;DR

This paper introduces LITE, a learnable late-interaction model for document ranking that achieves better accuracy, lower latency, and reduced storage compared to previous models by being a universal approximator of relevance scoring functions.

Contribution

LITE is a novel learnable late-interaction model that is theoretically a universal approximator and empirically outperforms existing models like ColBERT in relevance ranking tasks.

Findings

LITE outperforms ColBERT in MS MARCO passage re-ranking.

LITE achieves better generalization and lower latency.

LITE requires only 0.25x storage of ColBERT.

Abstract

Cross-Encoder (CE) and Dual-Encoder (DE) models are two fundamental approaches for query-document relevance in information retrieval. To predict relevance, CE models use joint query-document embeddings, while DE models maintain factorized query and document embeddings; usually, the former has higher quality while the latter benefits from lower latency. Recently, late-interaction models have been proposed to realize more favorable latency-quality tradeoffs, by using a DE structure followed by a lightweight scorer based on query and document token embeddings. However, these lightweight scorers are often hand-crafted, and there is no understanding of their approximation power; further, such scorers require access to individual document token embeddings, which imposes an increased latency and storage burden. In this paper, we propose novel learnable late-interaction models (LITE) that resolve these issues. Theoretically, we prove that LITE is a universal approximator of continuous scoring functions, even for relatively small embedding dimension. Empirically, LITE outperforms previous late-interaction models such as ColBERT on both in-domain and zero-shot re-ranking tasks. For instance, experiments on MS MARCO passage re-ranking show that LITE not only yields a model with better generalization, but also lowers latency and requires 0.25x storage compared to ColBERT.