The Evolution of Question Generation: Analyzing the Impact of Large Language Models

Abstract

Automatic question generation (AQG) has evolved quite a bit with the advent of transformer‐based large language models (LLMs), transitioning from rule‐based methods and early neural approaches to more adaptable transformer‐based and LLM‐based systems. This process has enabled greater potential for AQG but, at the same time, has posed new problems regarding control, structure, evaluation, and grounding of the generated questions in a wide range of fields. In this survey, we look at the evolution of AQG with respect to five different aspects—answer format, contextual scale, cognitive complexity, control, and multilingual or multimodal capacity. In our analysis of this transition, we focus on the progression of AQG from rule‐based and neural methods to transformer‐based and LLM‐based methods with an emphasis on prompting, retrieval‐augmented generation (RAG), and agentic workflows. Furthermore, we also discuss the technologies involved in modern AQG, such as LLMs, SLMs, retrieval, and agentic systems with reasoning and memory and tool‐use capabilities. Moreover, this paper contributes a taxonomy that captures paradigm shifts in model architecture, prompting, and pipeline design. It also compiles the existing datasets, evaluation techniques, and benchmarks, emphasizing the increasing relevance of retrieval‐based integration, controllable prompting, and agentic pipelines. These advances create ways for the generation of context‐ and cognition‐aware questions tailored to users' needs but also present new challenges associated with factual grounding and evaluation. In summary, the survey reflects the current trend toward developing more flexible, resilient, and multimodal AQG systems with wide‐ranging applications, including educational settings, knowledge‐based systems across various domains, and human‐computer collaboration. The paper helps establish a solid foundation for future research on question generation through its critical analysis of current trends and research directions.