본문 바로가기
  • Home

Improving Text-to-SQL Reliability by Leveraging Operational SQL

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(4), pp.147~156
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : February 2, 2026
  • Accepted : March 23, 2026
  • Published : April 30, 2026

Myoungkuk Nam 1 Namgyu Kim 1

1국민대학교

Accredited

ABSTRACT

Effective use of relational databases requires substantial SQL expertise, driving active research into Text-to-SQL, which aims to translate natural language queries into SQL statements. Recently, generation-based approaches leveraging large language models (LLMs) have gained considerable attention. However, purely LLM-based methods face inherent limitations in handling natural language ambiguity, schema linking, and the reliability of generated outputs. Against this backdrop, this study proposes OpSQL-Leverage (OSL), a methodology that exploits SQL queries accumulated in real-world operational environments as core knowledge assets. Specifically, OSL selects candidate SQLs from an operational database based on an LLM-generated SQL, then synthesizes these candidates with the original natural language query to produce the final SQL. Experiments on the SPIDER dataset show that OSL achieves higher accuracy than purely LLM-based Text-to-SQL models. This work is significant in that it extends retrieval-augmented generation (RAG) approaches to the Text-to-SQL domain by leveraging structured SQL assets within operational databases as external knowledge.

Citation status

* References for papers published after 2024 are currently being built.