Transformer and Large Language Models for Automatic Multiple-Choice Question Generation: A Systematic Literature Review

Developing multiple-choice questions manually requires a lot of time and effort. Automatic multiple-choice question generation is one of the solutions to alleviate the problem. The research in automatic multiple-choice question generation has been growing with the recent use of Transformer and Large-Language Models. However, existing literature reviews have not thoroughly covered the recent advances in methods and evaluation conducted on the multiple-choice question generation domain. This research conducted a systematic literature review on multiple-choice question generation using Transformer and Large Language Models. This research aims to discover recent methods and evaluation strategies that have been used in the domain. We obtained 28 primary studies. We presented a taxonomy covering strategy of using the Transformer and Large Language Models for multiple-choice question generation, including fine-tuning and prompt engineering with zero-shot, few-shot, chain-of-thought, and retrieval augmented generation. Primary studies used either or both automatic and manual evaluation for the generated questions from Transformer and LLM. We found that studies are still primarily in English, with few studies utilizing learning components such as learning objective, limited use of chain-of-thought, retrieval augmented generation, and open problem in automatic evaluation.