Development build. This page describes main, not a released package. The latest published Lodestar.Text is 0.6.0 — read its documentation.

Home › Text › Keyword search

Combining several rankings of the same documents into one.

public static class RankFusion

Example — two rankings that disagree, fused.

using Lodestar.Text.Search;

IReadOnlyList<SearchHit> fused = RankFusion.Rrf([[2, 0, 1], [0, 1, 2]]);

int best = fused[0].Document;  // => 0
int worst = fused[2].Document;  // => 1
double k = RankFusion.DefaultK;  // => 60

Remarks — reciprocal rank fusion, from Cormack, Clarke and Buettcher (2009). It is what makes a hybrid search: fuse a BM25 ranking with a vector one and neither side's scale has to be reconciled, because only positions are read. That is the whole reason to prefer it over normalising two score distributions and adding them.

A last place costs more than a first place gains. Document 2 is ranked first and then last; document 0 is ranked second and then first, and wins — 1/61 + 1/63 is less than 1/62 + 1/61. That asymmetry is the whole behaviour of k: at 60 the gap between consecutive ranks is small and consistency beats a single strong opinion, and lowering k reverses it.

There is no canonical Python library to check this against. It is one formula, so it is pinned by tests that state it — the same way Lodestar.Metrics' mean reciprocal rank is pinned — rather than by a frozen corpus. docs/equivalence.md records that exception instead of leaving the row looking skipped.

Thread-safe.

Applies to — net10.0, netstandard2.0.

See also — the search index, Bm25Index.Top.