# MCS-SQL: Leveraging Multiple Prompts and Multiple-Choice Selection For Text-to-SQL Generation

**Authors:** Dongjun Lee, Choongwon Park, Jaehyuk Kim, Heesoo Park  
**arXiv:** [2405.07467](https://arxiv.org/abs/2405.07467)  
**Date:** 2024-05-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2405.07467v1/

## Key points

- Study titled "MCS-SQL: Leveraging Multiple Prompts and Multiple-Choice Selection For Text-to-SQL Generation"
- Authors: Dongjun Lee, Choongwon Park, Jaehyuk Kim, Heesoo Park from Dunamu
- Introduces a novel approach leveraging multiple prompts to enhance search space for answers
- Key innovation in refining database schema through schema linking using multiple prompts
- Generates candidate SQL queries based on refined schema and diverse prompts
- Filters candidate queries based on confidence scores, selects optimal query through multiple-choice selection
- Achieves impressive execution accuracies of 65.5% and 89.6% on BIRD and Spider benchmarks respectively
- Surpasses previous ICL-based methods in accuracy
- Establishes new state-of-the-art performance on BIRD in terms of both accuracy and efficiency
- Promising approach to enhancing text-to-SQL generation by leveraging multiple prompts and incorporating sophisticated multiple-choice selection mechanism

## Summary

<MCS-SQL>, <Dongjun Lee>, <Choongwon Park>, <Jaehyuk Kim>, <Heesoo Park>

The study "MCS-SQL: Leveraging Multiple Prompts and Multiple-Choice Selection For Text-to-SQL Generation" by Dongjun Lee, Choongwon Park, Jaehyuk Kim, and Heesoo Park from Dunamu explores the advancements in large language models (LLMs) for text-to-SQL tasks. The researchers introduce a novel approach that leverages multiple prompts to enhance the search space for answers. The key innovation of the study lies in robustly refining the database schema through schema linking using multiple prompts. This process leads to the generation of various candidate SQL queries based on the refined schema and diverse prompts. These candidate queries are then filtered based on confidence scores, with the optimal query selected through a multiple-choice selection presented to the LLM. When evaluated on challenging benchmarks like BIRD and Spider, the proposed method achieves impressive execution accuracies of 65.5% and 89.6%, respectively, surpassing previous ICL-based methods. Additionally, the study establishes a new state-of-the-art performance on BIRD in terms of both accuracy and efficiency in generating queries. Overall, "MCS-SQL" presents a promising approach to enhancing text-to-SQL generation by leveraging multiple prompts and incorporating a sophisticated multiple-choice selection mechanism for improved query accuracy and efficiency.