BM25 Query Augmentation Learned End-to-End

TL;DR

This paper introduces an end-to-end learning approach to augment and re-weight BM25's query representation, significantly enhancing its retrieval performance while maintaining speed and demonstrating good transferability across datasets.

Contribution

It presents a novel method for learning query augmentation and re-weighting end-to-end, improving BM25's effectiveness without sacrificing efficiency.

Findings

Improved retrieval performance over standard BM25

Learned augmentations transfer well to unseen datasets

Retains BM25's speed despite enhancements

Abstract

Given BM25's enduring competitiveness as an information retrieval baseline, we investigate to what extent it can be even further improved by augmenting and re-weighting its sparse query-vector representation. We propose an approach to learning an augmentation and a re-weighting end-to-end, and we find that our approach improves performance over BM25 while retaining its speed. We furthermore find that the learned augmentations and re-weightings transfer well to unseen datasets.