13 Vector Embeddings

This chapter provides an overview of vector embeddings supported in Oracle® Database Navigator.

Database Navigator plug-in supports Vector data types that forms the basis of generating and storing vector embeddings in databases. Using embedding models, it is possible to transform unstructured data and data from tables into vector embeddings which can then be used for semantic search (instead of keyword-based queries) on business data.

Vector Embeddings are numerical representation of any kind of data (such as text, images, videos, audio, etc.) as points in a multi-dimensional space that describe the semantic meaning behind that data. The vector representation transforms the location of the data points in vector space (like coordinates) and their proximity to others, in a semantically meaningful form.

When performing a similarity search on an attribute of a data object, the vector equivalent of the search query is used to fetch the closest vector matches available for the data object. DBN utilizes Oracle's PL/SQL packages (DBMS_VECTOR and DBMS_VECTOR_CHAIN) to support operations with Oracle Vector AI Search, such as chunking the source data, extracting chunks or embeddings from the data, creating vector index, or processing the data for text generation and summarization, that helps in end-to-end semantic search.

For more information on Vector Embeddings and Oracle Vector AI Search, see Oracle AI Database Oracle AI Vector Search User's Guide, 26ai.

Topics: