SQLFixAgent: Towards Semantic-Accurate Text-to-SQL Parsing via Consistency-Enhanced Multi-Agent Collaboration

• DOI: 10.1609/aaai.v39i1.31979
• Publication Date: 2025-04-11T09:31:09Z
• Authors: Jipeng Cen; Jiaxin Liu; Zhixu Li; Jingjing Wang
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; Computation and Language (cs.CL)

While fine-tuned large language models (LLMs) excel in generating grammatically valid SQL Text-to-SQL parsing, they often struggle to ensure semantic accuracy queries, leading user confusion and diminished system usability. To tackle this challenge, we introduce SQLFixAgent, a new consistency-enhanced multi-agent collaborative framework designed for detecting repairing erroneous SQL. Our comprises core agent, SQLRefiner, alongside two auxiliary agents: SQLReviewer QueryCrafter. The agent employs the rubber duck debugging method identify potential mismatches between query. If error is detected, QueryCrafter generates multiple as candidate repairs using SQLTool. Subsequently, leveraging similar repair retrieval failure memory reflection, SQLRefiner selects most fitting statement from candidates final repair. We evaluated our proposed on five benchmarks. experimental results show that consistently enhances performance of baseline model, specifically achieving an execution improvement over 3% Bird benchmark. also has higher token efficiency compared other advanced methods, making it more competitive.