What is Embeddings

Definition

Embeddings are numerical vector representations of data such as words, sentences, images, or items, learned so that semantically similar things sit close together in vector space, allowing machines to measure similarity and meaning mathematically.

« Back to Glossary Index

Capture semantic meaning so similar items have similar vectors
Enable similarity search, clustering, and recommendation mathematically
Convert unstructured data into a form models and databases can compare
Transfer across tasks, providing reusable representations of content

Real World Example

A retailer generates embeddings for its products so that 'running shoes' and 'athletic trainers' land near each other in vector space, letting search and recommendations match by meaning rather than exact wording.

FAQs

What do embeddings represent?

They represent data as vectors in a space where proximity reflects semantic similarity between items.

How are embeddings created?

Models, often neural networks, are trained to map inputs to vectors such that related inputs produce nearby vectors.

Where are embeddings used?

In semantic search, recommendations, clustering, classification, and as the retrieval layer in RAG systems.