CHS-SQL: Confidence-Guided Heuristic Search for Schema Linking in Text-to-SQL
A new framework called CHS-SQL improves Text-to-SQL accuracy by balancing precision and recall during Schema Linking. The approach uses a heuristic search combined with model confidence to select candidate schema subsets, addressing the trade-off ignored by prior methods. It enables efficient fine-tuning of Small Language Models (SLMs) on a single NVIDIA RTX 4090 GPU, achieving performance close to large models while ensuring data security. The research, published on arXiv, demonstrates that both precision and recall of Schema Linking directly affect final SQL accuracy.
Key facts
- CHS-SQL is a novel framework for fine-tuning SLMs on Text-to-SQL tasks
- It balances precision and recall in Schema Linking
- Uses heuristic search combined with model confidence
- Runs on a single NVIDIA RTX 4090 GPU
- Achieves performance close to large models
- Ensures data security
- Published on arXiv with ID 2607.22624
- Schema Linking precision and recall directly affect SQL accuracy
Entities
Institutions
- arXiv