Retrieval Augmented Generation Based Thai Question-Answering System

Abstract

This paper proposes a Thai Question-Answering System, specifically developed by using the Computer Engi-neering curriculum at Srinakharinwirot University, Thailand. The proposed system integrates the OpenThaiGPT model within a Retrieval-Augmented Generation (RAG) framework to generate contextually accurate responses to user queries. To evaluate the system, a set of 41 questions are categorized into basic, intermediate, and advanced levels, employing both zero-shot and few-shot learning techniques to assess its performance. The experimental results show that few-shot prompting exhibits efficacy across most criteria and complex-ity levels. However, challenges in maintaining this efficacy were observed, particularly regarding the accuracy of responses. This highlights areas that require future improvement.