BM25 Query Augmentation Learned End-to-End
This work addresses the problem of boosting a widely used baseline in information retrieval, though it appears incremental as it builds directly on BM25.
The paper tackled improving BM25's information retrieval performance by learning to augment and re-weight its query vectors end-to-end, resulting in enhanced performance while maintaining speed and showing good transferability to unseen datasets.
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.