noun
An embedding is a numerical representation of data, like text or images, that captures its meaning in a way machines can process. It works by converting input into a vector—a list of numbers—where similar things are closer together in a multi-dimensional space.
Embeddings are often created using neural networks trained on vast datasets. For example, a language model like BERT generates embeddings by analyzing word relationships across millions of sentences, allowing it to group ‘king’ and ‘queen’ closer than ‘king’ and ‘apple.’
For users, embeddings change how AI tools work by enabling faster, more accurate searches and recommendations. They also let developers build smarter chatbots or image generators. However, creating high-quality embeddings can be costly—training a model like BERT typically requires thousands of dollars in cloud computing fees.
When you search for 'best hiking shoes' on an e-commerce site, embeddings help the AI match your query to products with similar descriptions.
In other languages
See also: Neural network