# Dense Retrieval

> Dense retrieval finds relevant passages by embedding the query and every passage as dense vectors, usually with a dual encoder, and returning the passages whose vectors are most similar to the query's.

- Identifier: PTL-0126
- Category: Retrieval & Tool Use
- Canonical URL: https://protologue.com/t/dense-retrieval/
- Also known as: semantic search, embedding-based retrieval, DPR, bi-encoder retrieval
- Introduced: 2020

## Description

Because it matches on meaning rather than shared words, dense retrieval finds paraphrases and related concepts that keyword search misses. It can miss exact identifiers, rare names, and codes, which is why production systems often combine it with BM25 in hybrid search and refine results with a reranker.

## Related terms

- [Embedding](https://protologue.com/t/embedding/)
- [BM25](https://protologue.com/t/bm25/)
- [Hybrid Search](https://protologue.com/t/hybrid-search/)
- [Reranking](https://protologue.com/t/reranking/)
- [Vector Database](https://protologue.com/t/vector-database/)

## Comparisons

- [Dense Retrieval vs BM25](https://protologue.com/compare/dense-retrieval-vs-bm25/)

## Sources

- Karpukhin et al. (2020). Dense Passage Retrieval for Open-Domain Question Answering. https://arxiv.org/abs/2004.04906

## Cite this entry

Protologue. (2026). Dense Retrieval. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0126). https://protologue.com/t/dense-retrieval/

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)