Embedding

An embedding turns an item such as a word, document, image, product, or user into a list of numbers that captures meaningful characteristics. Items with similar meanings or behavior are positioned near one another in this mathematical space, allowing software to compare them by distance rather than exact wording. A search system can therefore match a question with a relevant passage even when they use different phrases, while a recommendation engine can identify related products or interests. Embeddings are learned from data and reflect the strengths and biases of that data. Their usefulness depends on the model, dimensions, similarity measure, update strategy, and task, so an embedding suited to semantic search may not be appropriate for fraud detection or visual classification.