Text-to-SQL aims to translate natural language questions into executable SQL queries over structured databases, enabling non-expert users to access data intuitively. While recent advances in large language models (LLMs) have shown promise in this task, existing LLM-based approaches often struggle to strike a balance between strong reasoning capabilities and robust generalization.
To address these limitations, we propose CoTE-SQL to enhance the LLM-based text-to-SQL generation with three key innovations:
- Self-enhanced reasoning traces distilled from LLMs without human annotation;
- Structured chain-of-thought (CoT) prompting with modular decomposition and example retrieval;
- Error-aware revision based on SQL execution feedback.
Extensive experiments on the Spider and Bird benchmarks demonstrate that CoTE-SQL achieves new state-of-the-art performance among methods built on open-source LLMs with comparable model sizes, achieving 53.39% EX / 59.02% VES on Bird and 79.60% EX / 77.19% VES on Spider, with especially significant gains on complex queries.
Results highlight the effectiveness of combining self-enhancement, structured reasoning, and execution-time feedback within an LLM-based framework for text-to-SQL design.
Blogger's Review: This research successfully enhances the performance of text-to-SQL through innovative self-enhancement and structured reasoning methods, showcasing the potential of large language models in handling complex queries. The incorporation of execution feedback correction mechanisms further increases the system's practicality and intelligence, paving the way for future research.