ModelRefs / Dense Retrieval — AI Glossary
Dense Retrieval — AI Glossary
Retrieval using dense vector embeddings of queries and documents, finding semantically similar content via ANN search.
Overview
Dense retrieval (Karpukhin et al. 2020, DPR) encodes both queries and documents as dense vectors. Retrieval finds vectors closest to the query embedding using ANN indexes. Captures semantic similarity beyond keyword overlap. Requires embedding all documents at index time and a vector database at query time. Superior to sparse for paraphrastic and cross-lingual queries.
Reference details
| Topic | rag |
|---|---|
| Also known as | semantic retrieval, embedding retrieval |
| Last reviewed | 2026-06-24 |
Related terms
Example: The query that defeats it
Searching for error code “ORA-01555”, dense retrieval embeds it into a region of vector space near other database-error text and may return a passage about a different code entirely. BM25 matches the literal string and wins outright. This is precisely why production retrieval is usually hybrid: dense for paraphrase, sparse for identifiers.
Commonly confused with
Dense retrieval is not semantic search as a whole — it is one implementation of it. Semantic search is the goal; dense bi-encoder retrieval is the common technique. Sparse methods like SPLADE also target semantics while keeping a sparse representation.
When to use it
Reach for it when:
- Users paraphrase, and vocabulary between query and document differs
- Recall over a broad corpus matters more than exact matching
- You can afford an embedding step at both index and query time
Reach for something else when:
- Queries are identifiers, codes or exact names
- The domain vocabulary is far from the embedding model's training data
- You need to explain why a specific document was retrieved
Primary source
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Dense Retrieval — AI Glossary.
Frequently asked questions
What is Dense Retrieval?
Retrieval using dense vector embeddings of queries and documents, finding semantically similar content via ANN search.
Is Dense Retrieval the same as semantic retrieval?
Yes — semantic retrieval, embedding retrieval are common aliases for Dense Retrieval.
What concepts are related to Dense Retrieval?
Closely related concepts include sparse retrieval, bi encoder, embedding model.