Bilkent University
Department of Computer Engineering

---

CS 590/690 SEMINAR

 

Encoding Strategies in Text-to-SQL: A Performance Impact Study

 

Mousa Farshkar Azari
Ph.D. Student
(Supervisor: Prof.Dr.Özgür Ulusoy)
Computer Engineering Department
Bilkent University

Abstract: In this work, we explore the role of encoding strategies in the Text-to-SQL domain, focusing on the impact of different schema-linking mechanisms. Our study introduces a novel model and input encoding technique designed to enhance the efficiency of translating natural language questions into SQL queries. By applying our encoding strategies to the T5 transformer model, we categorize them into two distinct methods. This approach reveals that models with enriched inputs significantly outperform their counterparts without such enhancements. Our comparative analysis showcases the effectiveness of our method against existing encoding techniques, highlighting its potential to improve accuracy in the text-to-SQL task. The experiments are conducted across various datasets, demonstrating the versatility and robustness of our approach. Through this study, we contribute to a better understanding of how encoding strategies can influence the performance of text-to-SQL models. The findings suggest that careful consideration of input encoding mechanisms is crucial for advancing the state-of-the-art in this domain.

 

DATE: March 18, Monday @ 13:30 Place: EA 502