Text to SQL Generation Using Beam Search and an Enhanced Bert Model

Adhityasing Rajaput, Varsha Sangolli, Abhay Bellerimath, Amith Abbigeri, Uday Kulkarni, Shashank Hegde

2025

Abstract

The development of complex SQL queries is difficult for non-technical users, limiting access to valuable data for decision-making. This paper proposes a machine learning-based approach to generating Text-to SQL queries using a Transformer Encoder-Decoder architecture with a BERT model. The proposed system bridges the gap between natural language and structured query languages, thus allowing users to interact with databases through intuitive, natural language inputs. In addition to schema awareness and the efficiency of SQLgeneration, advanced techniques such as execution-guided decoding and beam search have been applied. Results for the model show query accuracy at eighty-four percent; the model is generally good at generat ing simple queries and basic aggregations, though complex and nested queries prove to be challenging to the model. This study focuses on the transformative nature of Text-to-SQL systems as means of improving access and efficiency in database interaction toward paving a future for improvement in conversational AI and smart management of databases. This model for Text-to-SQL query generation showed strong performance in con verting natural language queries to SQL commands. Its performance with regard to accuracy is close to about 92%, while BLEU stands at 0.78.

Paper Citation

in Harvard Style

Rajaput A., Sangolli V., Bellerimath A., Abbigeri A., Kulkarni U. and Hegde S. (2025). Text to SQL Generation Using Beam Search and an Enhanced Bert Model. In Proceedings of the 3rd International Conference on Futuristic Technology - Volume 3: INCOFT; ISBN 978-989-758-763-4, SciTePress, pages 49-55. DOI: 10.5220/0013608600004664

in Bibtex Style

@conference{incoft25,
author={Adhityasing Rajaput and Varsha Sangolli and Abhay Bellerimath and Amith Abbigeri and Uday Kulkarni and Shashank Hegde},
title={Text to SQL Generation Using Beam Search and an Enhanced Bert Model},
booktitle={Proceedings of the 3rd International Conference on Futuristic Technology - Volume 3: INCOFT},
year={2025},
pages={49-55},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013608600004664},
isbn={978-989-758-763-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 3rd International Conference on Futuristic Technology - Volume 3: INCOFT
TI - Text to SQL Generation Using Beam Search and an Enhanced Bert Model
SN - 978-989-758-763-4
AU - Rajaput A.
AU - Sangolli V.
AU - Bellerimath A.
AU - Abbigeri A.
AU - Kulkarni U.
AU - Hegde S.
PY - 2025
SP - 49
EP - 55
DO - 10.5220/0013608600004664
PB - SciTePress