Open Collections

Stenback, Soren

Text-to-SQL research has developed over several years to make use of Large Language Models (LLMs). One avenue of research is to create models that prompt an existing LLM such as ChatGPT and provide specific context that will assist the LLM in generating correct SQL queries based on a provided database schema and a text question. However, there are certain functions or techniques that do not assist the LLM and reduce the accuracy of the model. Researchers must determine which techniques are beneficial and which are detrimental to improving an LLM for text-to-SQL. The focus of this thesis is to assess an existing model called E-SQL and attempt to apply improvements through generative database descriptions to increase the accuracy of the model. Additional discoveries through testing are presented.

Text

University of British Columbia. COSC_O 449

Vancouver : University of British Columbia Library

10.14288/1.0448795

Science, Irving K. Barber Faculty of (Okanagan); Computer Science, Mathematics, Physics and Statistics, Department of (Okanagan)

Undergraduate

DSpace