Understanding Vector search in Azure AI Search

Understanding Vector Search in Azure AI Search

Vector search represents a significant advancement in information retrieval technology. Employing numeric representations of content, it allows for more flexible and semantically rich search experiences. In Azure AI Search, vector search is leveraged to enhance indexing, storage, and retrieval of vector embeddings. This high-level introduction explores the integration with Azure services, essential terminology, concepts, and practical steps to implement vector search in Azure AI Search.

Getting Started with Vector Search

To begin, generate vector embeddings or use integrated vectorization (preview). Add vector fields to an index, load vector data using push or pull methodologies, and query vector data using the Azure portal, REST APIs, or SDKs.

What is Vector Search?

Vector search in Azure AI Search allows for the indexing, storing, and retrieving of vector embeddings, powering similarity searches and recommendation engines. It utilizes a nearest neighbors algorithm to co-locate similar vectors.

Indexing and Query Workflows

The indexing side involves taking vector embeddings and using a nearest neighbors algorithm to group similar vectors. On the query side, the client application encodes the query input into a vector and sends it to the search index for similarity search.

Integration and Usage

Azure AI Search supports hybrid scenarios, allowing for indexing vector data alongside alphanumeric content. Vector queries can be combined with other query types, and the service offers flexibility in approach and integration with other Azure services.

Scenarios for Vector Search

Vector search can be used for text encoding, multimodal search, multilingual search, hybrid search, and filtered vector search. It also serves as a vector database for various applications.

Azure Integration and Related Services

Other Azure services like Azure OpenAI and Image Retrieval Vectorize Image API can provide embeddings. Azure AI Search can index vector data from Azure blob indexers and Azure Cosmos DB.

Core Concepts of Vector Search

Vector search is based on the concept of representing documents and queries as vectors. It uses machine learning models to generate these representations, allowing for semantic search across various media and languages.

https://learn.microsoft.com/en-us/azure/search/vector-search-overview