{
  "id": "embedding",
  "code": "PTL-0123",
  "term": "Embedding",
  "aliases": [
    "text embedding",
    "vector embedding",
    "embedding vector"
  ],
  "category": "retrieval-tools",
  "definition": "An embedding is a fixed-length vector of numbers that represents a piece of text, or another input, so that inputs with similar meaning have vectors that are close together, typically measured by cosine similarity.",
  "description": "Embeddings are produced by an embedding model, which is usually separate from the generating model. They underpin semantic search, dense retrieval for RAG, clustering, deduplication, and the selection of similar few-shot examples. Quality depends on the embedding model and on how text is chunked, and embeddings capture topical similarity better than exact facts such as numbers, names, or negation.",
  "example": null,
  "broader": [],
  "narrower": [],
  "related": [
    "dense-retrieval",
    "vector-database",
    "chunking",
    "retrieval-augmented-generation",
    "exemplar-selection"
  ],
  "introduced": 2019,
  "added": "2026-10-10",
  "sources": [
    {
      "title": "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
      "authors": "Reimers and Gurevych",
      "year": 2019,
      "url": "https://arxiv.org/abs/1908.10084"
    },
    {
      "title": "Dense Passage Retrieval for Open-Domain Question Answering",
      "authors": "Karpukhin et al.",
      "year": 2020,
      "url": "https://arxiv.org/abs/2004.04906"
    }
  ],
  "url": "https://protologue.com/t/embedding/",
  "citation": "Protologue. (2026). Embedding. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0123). https://protologue.com/t/embedding/"
}