LMAbsoluteDiscountingSimilarity

Namespace

Rowles.LeanCorpus.Search.Scoring

Assembly

Rowles.LeanCorpus.dll

Language-model similarity with absolute discounting smoothing. Subtracts a constant δ from each observed term count and redistributes the probability mass to unseen terms via the collection model.

Score = log((max(tf - δ, 0) + δ · |d| · P(t|C)) / |d|), where P(t|C) = collectionFrequency / totalTermsInCollection.

Default δ = 0.7.

public sealed class LMAbsoluteDiscountingSimilarity : ISimilarity

Inheritance

object

LMAbsoluteDiscountingSimilarity

Implements

ISimilarity

LMAbsoluteDiscountingSimilarity(float)

Initialises a new instance with the specified discount parameter δ.

Instance

Gets a shared singleton instance with the default δ = 0.7.

RequiresCollectionStatistics

Whether this similarity requires collection-level statistics (total term frequency and total terms in collection) for scoring.

PrecomputeFactors(int, int, float)

Precomputes factors constant for a given term across all documents.

PrecomputeLmFactors(int, int, float, long, long)

Precomputes factors including collection-level term statistics for language-model similarities. Default simply delegates to PrecomputeFactors(int, int, float) and sets CollectionProb to 0.

Score(int, int, float, int, int)

Computes the score for a single term occurrence in a document.

ScoreLmPrecomputed(float, float, float, int, int)

Scores using precomputed language-model factors. Default delegates to ScorePrecomputed(float, float, int, int), ignoring collectionProb.

ScorePrecomputed(float, float, int, int)

Scores using precomputed factors for hot-path scoring.