Skip to main content

Vector search

Advanced Managed Search supports high-dimensional embedding–based similarity search by using the vector search capability of OpenSearch. Beyond traditional keyword search, it enables proximity-based search using embedding vectors derived from text, images, audio, and other data.
Vector search can be used in various machine learning and AI–based services such as recommendation systems, semantic search, and natural language query processing.

What is a vector?​

A vector is a high-dimensional numeric array created using machine learning models or embedding techniques.
These vectors contain information about semantic similarity or distance between data, enabling similar data retrieval based on those relationships.

With vector search in Advanced Managed Search, you can use the following capabilities.

  • Similarity search

    • Finds the closest results by calculating distance or similarity between a query vector and stored data vectors.

  • High-dimensional space exploration

    • Supports searching using various distance or similarity metrics such as Euclidean distance and cosine similarity.

  • AI and machine learning integration

    • Semantic search using NLP and image embeddings
    • Candidate recommendation based on user–item similarity in recommendation systems

Vector search workflow​

Vector search generally follows this workflow.

  1. Generate data embeddings

    • Convert text, image, or audio data into vectors
    • Example: Convert a sentence into a 768-dimensional vector using an NLP model

  2. Index vectors

    • Index documents containing embedding vectors
    • Map the relevant field as a vector type for search

  3. Run vector query

    • Generate a query vector used as the search reference
    • Search for documents with the most similar vectors

  4. Return and post-process results

    • Perform post-processing such as sorting and filtering

Vector search use cases​

  • Semantic text search
  • Similar image search based on image embeddings
  • Recommendation filtering based on user behavior or preference similarity
  • Query expansion or natural language–based search

For more detailed usage methods and examples such as writing vector queries, mapping settings, and index structures, refer to the OpenSearch official documentation.