Research and Practice on Database Interaction Based on Natural Language Processing
Zeshun You, Jiebin Yao, Dong Cheng, Zhiwei Wen, Zhi-Liang Lu, Xianyi Shen · 2024
Data serves as the foundation for an enterprise's digital transformation, and its efficient utilization requires database support. SQL is highly complex and often considered unsuitable for non-technical users. Reducing the barriers to data utilization can be achieved by exploring natural language interactions with databases. The NL2SQL task aims to enable natural language interaction with databases. The emergence of large language models (LLMs) has spurred significant theoretical advances in NL2SQL, accelerating its development. However, an efficient NL2SQL system architecture has yet to be established. This paper presents an NL2SQL architecture that incorporates three methods of SQL generation. Additionally, five prevalent issues in NL2SQL systems are analyzed, and corresponding solutions are proposed.