What is an Embedding?

An embedding is a numerical vector that represents an item such as a word, sentence, image, product, or user in a high-dimensional space. Items with similar meaning or properties end up close to each other in that space, which makes embeddings the foundational data structure for semantic search, recommendation, and retrieval-augmented generation.

Definition

Embeddings are produced by neural networks that have been trained to map inputs to vectors in a way that preserves semantic relationships. A product embedding might encode style, category, price range, and customer demographics implicitly, without explicit feature engineering. Modern embedding models handle multiple modalities, so the same vector space can represent text descriptions, product images, and user behavior.

Why it matters

Embeddings turn fuzzy concepts such as similarity, relatedness, or intent into operations that databases can execute efficiently. Comparing two items reduces to computing the distance between their vectors. This makes it possible to surface relevant products from a vague query, group similar items for merchandising, or detect duplicate listings at catalog scale.

Use cases

In commerce, embeddings power semantic search, look-alike audiences, content-to-product matching for editorial pages, image search, and the retrieval step in RAG pipelines. They also support personalization by embedding both users and products in compatible spaces, so the next-best item can be selected through a vector lookup rather than a hand-written rule.

Related

Explore Agentic Frontend Management Platform · Personalization.