MongoDB

Database platform company. Relevant to search via MongoDB Atlas Search — a full-text and vector search capability built on Apache Lucene, embedded inside Atlas (MongoDB’s managed cloud database).

Search Contributions

Semantic Boosting — a two-phase hybrid retrieval technique named and described by Erik Hatcher: run vector search first, inject results as boost clauses into a final lexical $search query. Preserves native faceting, highlighting, and pagination.

Atlas Search — Lucene-powered full-text search integrated directly into the MongoDB document model. Supports $search (lexical), $vectorSearch (dense ANN), and $rankFusion / $scoreFusion fusion operators.

Fusion aggregation stages

StagePurpose
$rankFusionRRF over named input pipelines, with per-pipeline combination.weights
$scoreFusionRSF with input.normalization (none / sigmoid / minMaxScaler) and combination.method (avg or a custom expression)
$scorePlaces a computed value into $meta.score for pipelines that don’t already produce one; optionally normalizes
$meta: 'scoreDetails'Exposes the full per-pipeline computation — rank or raw score, weight, and contribution — for explainability

Because the 60 in the RRF denominator is a fixed built-in, a weight of 30 on each of two pipelines conveniently rescales the fused score into a ~0.0–1.0 range. $vectorSearch scores already arrive in 0.0–1.0, while lexical $search BM25 scores are unbounded — so Score Normalization matters far more on the lexical leg. Worked through in Reciprocal Rank Fusion and Relative Score Fusion.

People

Articles

The MongoDB Hybrid Search series by Erik Hatcher:

  1. Survey of the Hybrid Search Landscape
  2. Reciprocal Rank Fusion and Relative Score Fusion
  3. Hybrid Search Blueprint Series Semantic Boosting

Key Concepts