Dense Passage Retrieval Explainer
Dense passage retrieval represents a question and each candidate passage as compact vectors. A query encoder and a passage encoder are trained so relevant question-passage pairs have high similarity in the embedding space.
Document vectors are computed in advance and stored in a vector index. At search time, the query vector is compared with indexed vectors, often using approximate nearest neighbor search to return candidates quickly.
Dense retrieval can find semantically related passages even when the query and evidence use different words. It became an important component of open-domain question answering and retrieval-augmented generation.
The model may struggle with rare names, exact identifiers, new domains, or distinctions absent from its training data. Hybrid lexical retrieval, hard-negative training, reranking, and evaluation on realistic queries can improve reliability.
Dense passage retrieval encodes questions and passages as vectors and retrieves passages whose learned representations are close to the query.
Key terms include dense retrieval, dual encoder, query vector, passage vector, embedding, similarity, vector index, hard negative and reranking.
Separate encoders map queries and passages into representations that can be indexed in advance.
Nearest-neighbor search returns passages whose vectors are most similar to the query vector.
Identifiers, rare entities, and domain-specific terms may favor lexical signals.
Hybrid search and reranking can combine broad semantic recall with precise matching.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.
