Definition
Embeddings are numeric vector representations of text or other content, positioned so that semantically similar content ends up close together in that vector space, allowing similarity between a query and source documents to be measured mathematically rather than through exact keyword matching. They are the underlying representation a vector database stores and searches.
Key facts
- A mathematical representation (a vector of numbers) of meaning, not the text or content itself.
- Positions semantically similar content close together in vector space, enabling similarity search beyond exact keyword matching.
- The representation format a vector database is built to store and search efficiently.
- Quality of the embedding model directly affects how well semantically related content is actually matched during retrieval.
How does Embeddings work in maintenance?
Embeddings are vectors of numbers generated by a model to represent the meaning of a piece of text (or image, or other content), positioned in a high-dimensional space such that content with similar meaning ends up mathematically close together, regardless of whether the exact wording matches.
What does Embeddings require?
A vector database is the storage and search system built specifically to hold embeddings and quickly find the closest matches to a given query vector. Retrieval-Augmented Generation relies on embeddings and a vector database together as its retrieval mechanism, converting both the query and source documents into embeddings to find and retrieve the most relevant matching content.
FAQ
Frequently asked questions
How is content represented as vectors?
Through embeddings: numeric vector representations of text or content positioned so that semantically similar content is mathematically close together, enabling similarity search beyond exact keyword matching.
Last reviewed: