An information retrieval approach using dense vector representations (embeddings) to encode queries and documents. Unlike sparse methods like BM25, dense retrieval captures semantic meaning in continuous vector spaces, enabling neural search and forming the foundation of modern RAG systems.
Loading more......
Dense retrieval encodes queries and documents as dense vectors (embeddings) in a continuous high-dimensional space. Documents are retrieved based on vector similarity, capturing semantic meaning beyond keyword matching.
Seminal approach from Facebook AI (2020):
Best practice combines both:
Varies by embedding model and vector database platform.