OpenSearch:Data retrieval

Hybrid search combines traditional text search with vector search, leveraging both structured and unstructured data to enhance search result accuracy.

Vector search encompasses both dense and sparse vector retrieval. The overall score is calculated by summing the distances of dense and sparse vectors. In the case of Euclidean distance (SquareEuclidean), vectors that are closer together signify a higher similarity.

For text + vector hybrid searches, the text component's score is determined by the extent of keyword matching. A higher match equates to a higher score. The comprehensive final score combines the vector distance (lower is better) with the text score (higher is better), with a higher overall score indicating greater relevance. The OpenSearch Vector Search Edition allows for the adjustment of the impact of vector distance and text score by assigning different weights. For instance, decreasing the weight of vector distance while increasing the weight of text score can yield improved retrieval outcomes.

Vector search enables users to import generated vector data into the Vector Search Edition instance for vector search queries.

Single document multiple vector retrieval allows users to store various vector contents in a document's vector field for search purposes.

Multiple query retrieval permits the simultaneous submission of multiple query requests to enhance retrieval efficiency and response times.

Predictive query lets users generate vectors from text or images using the Vector Search Edition's built-in vectorization model and conduct retrieval queries via text or images.

Primary key query swiftly retrieves specific documents using unique identifiers, ensuring efficient data access.

Filter expression employs filter conditions to precisely narrow down query results and streamline the data screening process.

Inverted index utilizes an inverted index structure to expedite full-text indexing and rapidly identify relevant documents, thus enhancing query performance.