# Next-Generation Database Interfaces: A Survey of LLM-based Text-to-SQL

**Authors:** Zijin Hong, Zheng Yuan, Qinggang Zhang, Hao Chen, Junnan Dong, Feiran Huang, Xiao Huang  
**arXiv:** [2406.08426](https://arxiv.org/abs/2406.08426)  
**Date:** 2024-06-12  
**Summary page:** https://www.summarizepaper.com/en/paper/2406.08426v1/

## Key points

- The text-to-SQL task is complex due to difficulties in user question understanding, database schema comprehension, and SQL generation.
- Traditional text-to-SQL systems have relied on human engineering and deep neural networks for solutions.
- Pre-trained language models (PLMs) have shown promising performance but can struggle with the increasing complexity of modern databases and challenging user queries.
- Large language models (LLMs) are emerging as a potential solution to enhance natural language understanding in text-to-SQL tasks.
- LLM-based implementations offer unique opportunities and challenges that can significantly impact the field of text-to-SQL research.
- The authors provide an overview of current challenges in text-to-SQL, trace its evolutionary process, introduce datasets and metrics for evaluation, analyze recent advances in LLM-based techniques, discuss remaining challenges, and propose future research directions.

## Summary

In their paper titled "Next-Generation Database Interfaces: A Survey of LLM-based Text-to-SQL," authors Zijin Hong, Zheng Yuan, Qinggang Zhang, Hao Chen, Junnan Dong, Feiran Huang, and Xiao Huang delve into the challenges and advancements in generating accurate SQL from natural language questions. The <kw>text-to-SQL task</kw> is complex due to difficulties in user question understanding, database schema comprehension, and SQL generation. Traditional text-to-SQL systems have relied on human engineering and deep neural networks for solutions. However, with the rise of <kw>pre-trained language models (PLMs)</kw>, there has been a shift towards utilizing these models for text-to-SQL tasks. While PLMs have shown promising performance, they can struggle with the increasing complexity of modern databases and challenging user queries. This limitation has led to the need for more sophisticated optimization methods tailored to address these issues. The authors highlight the emergence of large language models (LLMs) as a potential solution to enhance natural language understanding in text-to-SQL tasks. By integrating LLM-based implementations, unique opportunities and challenges arise that can significantly impact the field of text-to-SQL research. In their comprehensive survey, the authors provide an overview of current challenges in text-to-SQL and trace its evolutionary process. They also introduce datasets and metrics designed for evaluating text-to-SQL systems before delving into a systematic analysis of recent advances in LLM-based text-to-SQL techniques. Furthermore, the paper discusses remaining challenges within this domain and proposes future directions for research. By exploring the capabilities of LLMs in natural language understanding within the context of text-to-SQL tasks, this survey offers valuable insights into potential solutions and innovations that could shape the future of database interfaces.