A Systematic Review of Automatic Neural Question Generation

The ability to formulate meaningful questions is a fundamental aspect of both human and artificial intelligence. Neural Question Generation (NQG) uses deep learning techniques to automatically generate relevant questions from a given context. NQG systems have significant applications in improving question-answering models, facilitating educational tools, and enhancing conversational agents such as chatbots. However, a key challenge in NQG is the effective selection of target sentences and concepts for question formulation. This paper presents a systematic literature review (SLR) of NQG, analyzing different datasets, input preprocessing methods, methodologies, and evaluation techniques. We also highlight emerging trends and future directions in the field. Our review provides a comprehensive overview of NQG research, offering insights into current progress and remaining challenges. We find that all NQG models share a common Seq2Seq framework. In addition, the integration of Seq2Seq with attention mechanisms, as well as the use of part-of-speech (POS) tagging and named entity recognition (NER), contributes to the generation of accurate questions.