Dense Retrieval
Retrieval by embedding meaning into vectors and finding nearest neighbors — matches on semantics, not keywords.
Dense retrieval finds relevant documents by encoding both query and documents into dense embedding vectors and returning the nearest ones — matching on meaning. It is the “dense” counterpart to sparse, keyword-based methods like BM25.
Worked example: a query “how to reset my password” retrieves a doc titled “account recovery steps” because their embeddings are close, even with no shared keywords — something exact-match search would miss. Gotcha: dense retrieval can miss exact terms (a specific SKU, error code, or rare name) that sparse methods nail, and it depends heavily on the embedding model matching your domain — which is why production systems usually combine dense and sparse into hybrid search.