RESEARCH ON AN AUTOMATIC MULTIPLE-CHOICE QUESTIONS GENERATION METHOD

Abstract

The multiple-choice test format is becoming widespread for its convenience. However, manual multiple-choice questions (MCQs) generation is time-consuming and costly. Therefore, automatic MCQs generation from texts has become a popular research area. Along with the growth of artificial intelligence (AI) in general and natural language processing (NLP) in particular, many large language models (LLMs) were developed with the ability of understanding text and processing information in high accuracy. Taking those advantages, this paper proposes a method on automatic MCQs generation using popular LLMs, ChatGPT and Gemini, in combination with a technique that has never been applied to this domain - ReAct Agent. We evaluated the effectiveness of the proposed method by generating questions in Vietnamese for Operating System course of Posts and Telecommunications Institute of Technology. The conducted experiments shows that our method achieved the accuracy of 89\%, a promising result to apply on other courses.