Skip to main content
COMPEL Body of Knowledge
Get Certified Platform

COMPEL Glossary / hybrid-retrieval

Hybrid retrieval

A retrieval pattern combining dense (vector-based, e.g., dense passage retrieval) and sparse (term-based, e.g., BM25) retrieval methods, whose candidate sets are fused — typically via reciprocal rank fusion — before reranking.

What this means in practice

Produces consistently better recall than either method alone because the two approaches fail in different ways.

Synonyms

dense-sparse retrieval , fusion retrieval , hybrid search

Related Terms Network

This diagram shows how Hybrid retrieval connects to other concepts in the COMPEL body of knowledge. Each linked node is a term you can explore in turn.

Related terms for Hybrid retrieval Network graph showing 4 related glossary terms connected to Hybrid retrieval. Hybrid retriev… Reranker Naive RAG Embedding … Index algo…
4 related concepts connected to Hybrid retrieval.

The COMPEL Glossary Graph visualizes relationships between framework terminology, showing how concepts interconnect across domains, stages, and pillars. Term nodes cluster by pillar affiliation while cross-references reveal semantic dependencies — for example, how risk appetite connects to control effectiveness, model governance, and assurance requirements. This network representation helps practitioners navigate the framework vocabulary and understand that COMPEL terminology forms a coherent conceptual system rather than isolated definitions.

Accessible list of related terms
  • Reranker — A cross-encoder model that re-scores a candidate set of documents — typically the top-k from an initial vector or hybrid retrieval — to improve ordering by the actual query-document semantic match.
  • Naive RAG — The baseline retrieval-augmented-generation pattern: a single retrieval step returns the top-k chunks by vector similarity, and those chunks are concatenated into the prompt for generation.
  • Embedding model — A model that maps text, images, or multimodal content to dense vector representations used for retrieval, clustering, and similarity search.
  • Index algorithm — The data-structure choice — HNSW, IVF, product quantization (PQ), or flat (brute-force) — that governs a vector store's query latency, memory footprint, and recall-at-k tradeoff.

See also

  • Reranker — A cross-encoder model that re-scores a candidate set of documents — typically the top-k from an initial vector or hybrid retrieval — to improve ordering by the actual query-document semantic match.
  • Naive RAG — The baseline retrieval-augmented-generation pattern: a single retrieval step returns the top-k chunks by vector similarity, and those chunks are concatenated into the prompt for generation.
  • Embedding model — A model that maps text, images, or multimodal content to dense vector representations used for retrieval, clustering, and similarity search.
  • Index algorithm — The data-structure choice — HNSW, IVF, product quantization (PQ), or flat (brute-force) — that governs a vector store's query latency, memory footprint, and recall-at-k tradeoff.

Related articles in the Body of Knowledge