Stockertim Dictionary · Artificial intelligence · Technology
Embedding
Also called embeddings, text embedding, vector embedding
In short
An embedding is a long list of numbers that represents the meaning of a piece of text, so that texts with similar meanings end up with similar lists even when the words differ.
- Advanced
- Technology
- Checked
In detail
An embedding model reads a passage and outputs a fixed-length list of numbers, often hundreds or thousands long. Each list can be pictured as a point in a space with that many directions. The model is trained so that passages with related meaning land close together: a question about annual leave sits near a policy paragraph on holiday entitlement, and far from one on parking.
Closeness is measured mathematically, most often with cosine similarity, which compares the direction of two lists. A score near 1 means very similar meaning; a score near 0 means unrelated. Because the comparison is arithmetic, a system can compare a question with millions of stored passages quickly and pick out the nearest few.
Embeddings power semantic search, retrieval-augmented generation, duplicate detection and grouping of similar documents. They also carry the limits of the model that made them: they can miss domain jargon, reflect biases in the training text and treat negation loosely, so retrieved results are still checked.
Worked example
A simplified two-number illustration: a question has the embedding (0.8, 0.6); document A has (0.6, 0.8) and document B has (1.0, 0.0). All three lists have length 1, so cosine similarity is just the sum of the paired products.
- Question and document A: (0.8 x 0.6) + (0.6 x 0.8) = 0.96
- Question and document B: (0.8 x 1.0) + (0.6 x 0.0) = 0.8
Document A scores 0.96 against 0.8, so it is retrieved first as the closer match in meaning.
In practice
Developers embed documents once, store the lists in a vector database and embed each new question to find related passages; data teams also use embeddings to cluster feedback and find duplicates.
Watch out
- An embedding represents meaning statistically; it is not a summary or a translation of the text.
- Embeddings from different models are not comparable with each other.
Learn it properly
Official sources
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile · National Institute of Standards and Technology
- AI data security: best practices for securing data used to train and operate AI systems · Cybersecurity and Infrastructure Security Agency
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.