kandpal-2023-bm25-doc-count-bins-range
IN premise — summaries/2026/08/24/kandpal-2023-long-tail-knowledge-s5-related-work.md
Created 2026-08-25T02:58:07+00:00
The BM25 retrieval augmentation evaluation reports accuracy across document-count bins ranging from 10^0 to 10^6 relevant documents.
Summary
The BM25 retrieval evaluation in question measures accuracy across a very wide range of knowledge-base sizes, from a single relevant document up to one million, so its results are intended to be meaningful whether you are searching a tiny corpus or a massive one. This matters because any conclusion drawn from that evaluation only covers that span of scale, and results outside it would need separate validation.