Google Cloud Adds Native BM25 Ranking to AlloyDB and Cloud SQL for PostgreSQL 17+

Native BM25 Ranking Introduced

Google Cloud has announced the preview of native BM25 (Best Matching 25) ranking within AlloyDB and Cloud SQL for PostgreSQL 17+. This new capability integrates industry-standard keyword ranking directly into these database services, which previously required an additional full-text search backend.

Addressing Hybrid Search Needs

Vector search is a key component for generative AI and retrieval-augmented generation (RAG), but it struggles with specific alphanumeric IDs and exact product SKUs. The addition of native BM25 ranking allows for a unified hybrid search approach, combining semantic vector search with traditional exact keyword full-text search, which is necessary for robust AI applications.

Operational Simplification

The integration of BM25 ranking directly into the database eliminates data silos, synchronization lags, and operational complexity associated with maintaining separate full-text search backends. Users no longer need to provision, manage, or pay for distinct systems to achieve state-of-the-art full-text retrieval, as it now occurs within the database where operational data resides.

Powered by pg_textsearch

This new functionality is made possible through the open-source pg_textsearch extension, created by TigerData. Unlike PostgreSQL's built-in ts_rank, BM25 provides improved ranking quality by supporting inverse document frequency and term-frequency saturation, which are crucial for effective information retrieval at scale.

Performance Benefits

The native BM25 scoring is C-optimized for fast performance directly on Postgres tables. Additionally, AlloyDB users can experience up to 6x and 10x faster vector search queries compared to standard PostgreSQL, thanks to ScaNN and HNSW index types, further enhancing semantic search capabilities.