Amazon S3 Vectors adds metadata pre-filtering capability
Amazon has introduced metadata pre-filtering for S3 Vectors, allowing users to filter attributes, such as category or status, before running a similarity search. This approach increases the number of relevant results for queries limited by specific metadata. Filters can now use prefix matching for fields like paths and URLs. The update requires no additional cost or data re-ingestion, and aims to improve recall in filtered semantic searches and retrieval-augmented generation applications.
- Pre-filtering evaluates metadata before similarity search
- Supports prefix matching for paths and hierarchical keys
- No extra cost or need to re-ingest data
- Up to 2 KB metadata per vector, 100 filters per query
- Improves recall in filtered vector search scenarios
Sources covering this
In this story
More in Enterprise
Amazon S3 Tables add full Apache Iceberg V3 support
Amazon's S3 Tables now fully support all data types from the latest Apache Iceberg V3 standard.
Microsoft adds native Linux containers to Windows 11
Microsoft has made it possible to run Linux containers directly on Windows 11, without the usual third-party workarounds.
Amazon Aurora PostgreSQL now queries Iceberg, Parquet data lakes
Amazon Aurora PostgreSQL just rolled out a feature that lets you run SQL queries on both your live operational data and big-data files…
Google Cloud previews remote CLI server for AI agents
Google Cloud has launched a preview of its CLI remote MCP server.