Word embeddings are numeric representations of words that capture meaning and relationships in a continuous vector space. Each word maps to a list of real-valued numbers, typically 50 to 300 dimensions. Words that appear in similar contexts end up close to each other. The vectors are learned from large text corpora and reflect statistical patterns of language without hand-crafted rules. The classic example is the king-queen analogy. In embedding space, the vector from "king" to "queen" is roughly the same as the vector from "man" to "woman." The algorithm discovers these relationships automatically by reading millions of sentences. Word embeddings matter because they turn messy text into linear algebra that computers can process. Models using vectors perform classification, translation, sentiment analysis, and question answering far more efficiently than models manipulating raw strings. Embeddings power search engines that rank by semantic similarity, voice assistants that interpret user intent, and recommendation systems that match products to user reviews. Libraries like GloVe, fastText, and BERT produce reusable embeddings that accelerate both research and production systems.
Back to Glossary