Searchplex ID Retrieval terminology registry

Representation

Embedding

Canonical ID https://id.searchplex.net/embedding/

A vector representation of an item, token, query, document, or other object in a continuous or otherwise structured representation space.

Also called

  • Vector Embedding

Scope

In retrieval systems, embeddings are commonly used as vector representations for similarity search, dense retrieval, reranking, clustering, or generation support. An embedding is the representation object, not the model that produced it and not the retrieval system that searches over it.

Evaluated by

Not equivalent to

  • Embedding is not Dense Encoder. A dense encoder is a component or model that produces dense representations; an embedding is the produced representation.
  • Embedding is not Dense Retrieval. An embedding is a representation; dense retrieval is a retrieval approach that may use dense embeddings.
  • Embedding is not Vector Search. Vector search retrieves over vector representations; an embedding is one such representation.
  • Embedding is not Chunking. Chunking decides the units to represent; embedding is the representation of a unit.
  • Embedding is not MTEB. MTEB is a benchmark suite for evaluating embedding models; an embedding is a vector representation.
  • Embedding is not Vector Space Model. The vector space model is a retrieval model over term-weighted vectors; an embedding is a learned vector representation that may be used in many retrieval or non-retrieval settings.

Learn more

ID