Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

CRED-SQL: Enhancing Real-world Large Scale Database Text-to-SQL Parsing through Cluster Retrieval and Execution Description

Created by
  • Haebom

Author

Shaoming Duan, Zirui Wang, Chuanyi Liu, Zhibin Zhu, Yuhao Zhang, Peiyi Han, Liang Yan, Zewu Peng

Outline

This paper proposes the CRED-SQL framework to improve the accuracy of Text-to-SQL systems, which convert Natural Language Queries (NLQs) into SQL queries in large-scale databases. Existing Text-to-SQL systems suffer from poor accuracy due to schema matching errors and semantic drift caused by semantically similar attributes in large databases. CRED-SQL resolves this schema mismatch problem by accurately identifying tables and columns related to NLQs through cluster-based, large-scale schema search. Furthermore, by introducing Execution Description Language (EDL), an intermediate representation language between NLQ and SQL, CRED-SQL decomposes the task into two steps: Text-to-EDL and EDL-to-SQL. This decomposition leverages the powerful inference capabilities of LLMs while reducing semantic drift. Experimental results on two large-scale cross-domain benchmarks, SpiderUnion and BirdUnion, demonstrate CRED-SQL's effectiveness and scalability by achieving state-of-the-art performance.

Takeaways, Limitations

•
Takeaways:
◦
We propose a new framework, CRED-SQL, that significantly improves the accuracy of Text-to-SQL systems in large-scale databases.
◦
Addressing schema mismatch and semantic drift issues through cluster-based schema discovery and EDL intermediate representation language.
◦
Achieving state-of-the-art performance in two large-scale benchmarks.
◦
Ensure reproducibility and extensibility through open code.
•
Limitations:
◦
Further research is needed to determine the generalization performance of the proposed method. It is also necessary to verify its dependence on specific database structures or query types.
◦
Further research is needed to optimize EDL design and improve the efficiency of the EDL-to-SQL conversion process.
◦
Further performance evaluations are needed for databases of varying size and complexity.
👍
Made with Slashpage