Module hybrid 

Hybrid search combining dense and sparse retrieval. Hybrid search combining dense and sparse retrieval.

This module provides fusion algorithms and search orchestration for combining dense semantic search (HNSW) with sparse keyword search (BM25).

Fusion Methods

Two fusion methods are supported:

1. Reciprocal Rank Fusion (RRF): Combines ranks using score(d) = sum(1 / (k + rank_i(d))). Robust, parameter-insensitive.

2. Linear Combination: Combines normalized scores using score(d) = alpha * dense_score + (1 - alpha) * sparse_score.

Example

use edgevec::hybrid::{rrf_fusion, FusionResult};

// Dense search results: [(id, score), ...]
let dense = vec![(1, 0.95), (2, 0.80), (3, 0.75)];

// Sparse search results: [(id, score), ...]
let sparse = vec![(2, 5.5), (4, 4.2), (1, 3.8)];

// Fuse with RRF (k=60)
let fused = rrf_fusion(&dense, &sparse, 60, 10);

// Results combine ranks from both lists
for result in &fused {
    println!("ID: {}, RRF Score: {}", result.id, result.score);
}

Structs

FusionResult

Result from fusion algorithm.

HybridSearchConfig

Configuration for hybrid search.

HybridSearchResult

Result from hybrid search.

HybridSearcher

Hybrid search combining dense and sparse retrieval.

Enums

FusionMethod

Fusion method configuration.

HybridError

Errors that can occur during hybrid search.

Constants

RRF_DEFAULT_K

Default k parameter for RRF fusion.

Functions

linear_fusion

Linear combination fusion algorithm.

rrf_fusion

Reciprocal Rank Fusion (RRF) algorithm.