Lodestar.Text 0.6.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

Scores every document against a query, in document order.

public double[] Score(IEnumerable<int> queryTerms)

Parameters — queryTerms are column indices, one per query occurrence. A term given twice counts twice; an index outside the vocabulary is ignored.

Returns — double[], one score per document, in the matrix's row order.

Exceptions — ArgumentNullException when queryTerms is null.

Example — a term in every document, where the score is small rather than large.

using Lodestar.Abstractions;
using Lodestar.Text.Search;
using Lodestar.Text.Vectorization;

var vectorizer = new CountVectorizer();
CsrMatrix counts = vectorizer.FitTransform(
    ["the cat sat", "the dog sat sat", "a bird flew far away today"]);
int sat = vectorizer.GetFeatureNames().ToList().IndexOf("sat");

double[] scores = new Bm25Index(counts).Score([sat]);

double shorter = scores[0];  // => 0.0799…
double twice = scores[1];  // => 0.1013…
double absent = scores[2];  // => 0

Remarks — sat appears in two of the three documents, so Robertson's IDF for it is negative and gets floored; the surviving weight is small, which is the point. Compare cat, in one document of three, which scores 0.5755… — several times more for the same single occurrence.

The second document holds sat twice and scores more than the first, but not twice as much: that is K1 saturating the term frequency. It is also longer, which B penalises — the two pull against each other and the parameters decide by how much.

A score can be negative, and Score does not clamp. It happens when the floored IDF is itself negative, which is what Bm25Idf.RobertsonFloored does in a corpus whose mean raw IDF is below zero. Clamping would hide a genuine property of the weighting.

Applies to — net10.0, netstandard2.0.

See also — Bm25Index, Bm25Index.Top.