What is a vector embedding?

A vector embedding is a numeric representation of an entity (such as a word, sentence, image, or user) in a continuous vector space. Each dimension of the vector captures latent features learned from data. Embeddings allow us to compute similarity, perform clustering, and feed them into machine learning models. For example, the word "king" might be represented as a 300‑dimensional vector that is close to "queen" and far from "car". Embeddings are learned by training models on large corpora, optimizing for tasks like predicting context words or reconstructing input. They reduce high‑dimensional categorical data to a compact, dense format that preserves semantic relationships. Because embeddings are dense, they are memory‑efficient and enable fast vector operations, making them ideal for real‑time recommendation and search.