Quality and Cost Trade-offs in Passage Re-ranking Task

TL;DR

This paper explores how to balance the quality and computational cost of passage re-ranking in information retrieval by evaluating various transformer-based models and techniques to optimize efficiency without significantly sacrificing accuracy.

Contribution

It introduces a systematic analysis of different re-ranking architectures and methods, including binarization, to improve efficiency in real-time retrieval systems.

Findings

Binarization of transformer outputs reduces memory footprint.

Late-interaction models like Colbert and Poly-encoder perform well with optimized configurations.

Trade-offs between model size, speed, and ranking quality are quantified.

Abstract

Deep learning models named transformers achieved state-of-the-art results in a vast majority of NLP tasks at the cost of increased computational complexity and high memory consumption. Using the transformer model in real-time inference becomes a major challenge when implemented in production, because it requires expensive computational resources. The more executions of a transformer are needed the lower the overall throughput is, and switching to the smaller encoders leads to the decrease of accuracy. Our paper is devoted to the problem of how to choose the right architecture for the ranking step of the information retrieval pipeline, so that the number of required calls of transformer encoder is minimal with the maximum achievable quality of ranking. We investigated several late-interaction models such as Colbert and Poly-encoder architectures along with their modifications. Also, we took care of the memory footprint of the search index and tried to apply the learning-to-hash method to binarize the output vectors from the transformer encoders. The results of the evaluation are provided using TREC 2019-2021 and MS Marco dev datasets.