Embedding
Also: Vector embedding, Vector representation, Feature vector, Dense vector
An embedding is a numerical vector that represents data (text, images, or items) in a way that captures meaning, so similar items sit close together in space.
What it is
An embedding is a list of numbers (a vector) that represents a piece of data such as a word, sentence, image, product, or user. Instead of treating data as raw symbols, a model maps it into a continuous vector space where distance reflects similarity. Items with related meaning end up close together, while unrelated items sit far apart.
For example, the words "car" and "automobile" would have nearby vectors, while "car" and "banana" would be far apart. The vector itself might have hundreds or thousands of dimensions (for instance 384, 768, or 1536 values).
Why it matters
Computers cannot directly compare meaning in raw text or pixels. Embeddings turn unstructured data into a structured numerical form that algorithms can measure, search, and compare.
- Semantic search: find results by meaning, not just keyword match.
- Recommendations: suggest items similar to what a user liked.
- Clustering and classification: group or label data automatically.
- Retrieval for AI: feed relevant context to large language models (RAG).
This makes embeddings a foundation of modern AI systems across data, marketing, and finance.
How it is used in practice
1. Generate: pass data through an embedding model to get vectors.
2. Store: save vectors in a vector database (such as one supporting indexes for fast lookup).
3. Compare: measure closeness using cosine similarity or Euclidean distance.
4. Act: return the nearest neighbors as search results, recommendations, or context.
Similarity is often computed with cosine similarity, which looks at the angle between two vectors rather than their length.
Concrete example
A marketing team has 50,000 support tickets. They embed each ticket into a vector, then cluster the vectors. Tickets about "billing errors" naturally group together even when customers used different wording ("wrong charge", "overbilled", "incorrect invoice"). The team discovers a recurring problem without reading every ticket.
In a finance setting, embeddings of company filings let analysts retrieve passages that discuss "liquidity risk" even when those exact words are absent, surfacing relevant disclosures faster.
Key points
- Embeddings capture meaning, not exact spelling.
- Quality depends on the model used to create them.
- They power search, recommendations, clustering, and AI retrieval.