Text-to-SQL framework learns database structure from query logs
Original title: Schema-Free Text-to-SQL: Learning Enterprise Database Structure from Query Logs
On Beaver it reports 95.7% execution validity versus 60.8% for embedding retrieval, with experts preferring it 70% to 10%.
The paper presents a text-to-SQL framework that skips schema documentation and learns structure from historical query logs. It builds a knowledge graph from parsed query patterns and uses entity-driven GraphRAG retrieval to ground questions in domain terms. On Beaver, the authors report 95.7% execution validity against 60.8% for embedding-based retrieval. In an internal deployment, expert data engineers preferred the system 70% to 10%.
- The method targets databases with cryptic names and missing documentation.
- Query logs supply co-queried tables, joins, and column-to-concept mappings.
- Evaluation used Beaver plus expert review of an industrial deployment.
- Authors say it beat similarity retrieval on executability and answer quality.
From the original article
1 min read at arxiv.orgText-to-SQL systems achieve remarkable accuracy on academic benchmarks, yet performance drops dramatically in enterprise deployments. This gap stems from a fundamental assumption: existing approaches require complete schema documentation, but enterprise databases typically lack comprehensive documentation, featuring cryptic naming and implicit relationships.
By Ömer Faruk Akgül · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.