Enable Metadata Filter Queries for OpenSearch

[philikai]

Enable Metadata Filter Queries for OpenSearch:

Implement Approximate KNN Search with Metadata Filtering in Amazon OpenSearch

Goal

• Implement self-querying or metadata filtering for approximate KNN search in Amazon OpenSearch

Motivation

• Customers often need to filter based on metadata depending on the use case. E.g. filter for a certain topic or business unit
• More data in one index makes it harder to match on semantic meaning as well as filter e.g. for entries that have been added on a specific date.
• Matching on metadata can be helpful to improve query performance as well as retrieval performance.

Approaches

UI Integration

• Allow end user to specify filters on metadata for an OpenSearch index

Query Rewriting by LLM

• Have the LLM itself rewrite the query before search execution
• Langchain self-query retriever is an example of how to do this https://python.langchain.com/docs/modules/data_connection/retrievers/self_query/

[ystoneman]

OpenSearch 2.9 on Amazon OpenSearch Service offers efficient vector query filtering with FAISS for fast metadata filtering during approximate k-NN search.

UI integration may best expose filters to users, while LLM query rewriting could enable natural language filtering to complement the UI.

Key questions:

• Which metadata fields should be filterable?
• How can the UI support both basic and advanced filtering?
• What performance issues emerge with heavy filtering?
• Should we start with one approach or both?
• Your thoughts on these questions and the direction are valuable. A PR would be welcome if you'd like to contribute. Let's refine the design and implementation together.

Share ideas or open a draft PR for feedback. Excited to see this progress!

[github-actions[bot]]

This issue is stale because it has been open for 60 days with no activity.

[github-actions[bot]]

This issue was closed because it has been inactive for 30 days since being marked as stale.