dense retrieval

Finding relevant text by nearest-neighbor search over embedding vectors rather than exact keyword matching.

Dense retrieval finds relevant passages by embedding a query and a corpus of documents into the same vector space, then returning the documents whose embeddings are nearest the query's embedding by cosine similarity -- "dense" contrasts with sparse, keyword-based methods like TF-IDF or BM25, which represent text as mostly-zero count vectors over a huge vocabulary.

Examples

Dense retrieval can match the query "how do I get my money back" to a document titled "Refund Policy" even though the two share no words in common.