Abstract:
The field of database management often presents challenges to individuals lacking the technical expertise necessary for constructing and executing SQL queries. Implementing text-to-SQL systems, which translate user-friendly natural language queries into structured SQL commands, is a viable solution to this problem. This thesis investigates the creation of a novel text-to-SQL system that addresses the deficiencies identified in the current landscape of such translation systems. The primary objective of this research is to address situations in which lengthy or complex natural language descriptions cannot be accurately converted to SQL queries. This study presents a novel combination of a pre-trained RoBERTa model and a relation-aware transformer network in order to bridge this divide. This merger seeks to improve translation accuracy by combining the rich linguistic understanding of RoBERTa with the relation-aware transformer's ability to comprehend the database schema.
