Redis
redis-vl-dotnet Documentation redis-vl-dotnet

redis-vl-dotnet

HybridSearchQuery

HybridSearchQuery runs a native FT.HYBRID query (Redis 8.4+). The text branch (SEARCH) and the vector branch (VSIM) are scored independently and fused server-side, giving you tunable weighting between lexical and semantic relevance.

FT.HYBRID requires Redis 8.4 or newer. On older servers use HybridQuery (a single FT.SEARCH expression) or AggregateHybridQuery (FT.AGGREGATE) instead.

Basic shape

var query = HybridSearchQuery.FromFloat32(
    Filter.Text("title").Prefix("He"),   // SEARCH branch (must contain a text predicate)
    "plot_embedding",                     // VSIM branch
    queryVector,
    topK: 3,
    combination: new LinearHybridCombination(alpha: 0.7, beta: 0.3),
    vectorFilter: Filter.Tag("genre").Eq("crime"),
    returnFields: ["title"]);

var results = await index.SearchAsync(query);
var typed = await index.SearchAsync<Movie>(query);

Each returned document exposes the fused score under HybridSearchQuery.ScoreField (__score); the source document key becomes the result Id.

Fusion strategies

The combination argument controls how the two branches are merged:

  • LinearHybridCombination(alpha, beta, window?) — weighted blend where alpha weights the text score and beta weights the vector score (COMBINE LINEAR).

  • ReciprocalRankFusionHybridCombination(constant?, window?) — reciprocal rank fusion (COMBINE RRF). At least one of constant or window must be supplied.

  • null (default) — use the server default, which is RRF with a window of 20 and a constant of 60.

Filtering

  • The textQuery (the first argument) drives the lexical SEARCH branch and must contain at least one text predicate, otherwise construction throws ArgumentException.

  • vectorFilter applies an optional pre-filter to the vector branch (VSIM …​ FILTER), restricting the candidate set considered during nearest-neighbour search.

Paging and runtime behavior

  • topK (the KNN K count) must be greater than zero.

  • pagination maps to LIMIT offset num and defaults to the first topK results.

  • VectorKnnRuntimeOptions.EfRuntime emits EF_RUNTIME for HNSW vector fields.

Return-field normalization and typed mapping behave the same way as VectorQuery.

Related pages