Predicting learners' multi-question performance based on neural networks

As massive open online courses (MOOCs) and online intelligent tutoring systems(ITS) have become increasingly widespread, the number of learners enrolled in online courses has shown explosive growth. However, these learners are likely to have acquired knowledge from diverse educational and vocational backgrounds. Therefore, it is unwise to apply the same criteria and assessment questions to assess all learners' abilities without differentiation. Therefore, the demand for the adaptive arrangement of questions for online learners is ever critical. Deep learning is a new increasingly popular approach for handling extraordinarily complex problems such as image recognition and natural language processing. In this research, we use neural networks to forecast learners' multi-question performance on new test questions and propose a new concept called predictable property for the first time to explain the reasons why neural networks can be applied to predict learners' multi-question performance based on their previous question responses. This approach means that fewer questions need to be answered by learners although more information can be gathered about them through the use of deep-learning-based techniques. Finally, we use both artificial datasets generated by cognitive models and three real-world datasets to validate the algorithm's performance. Experiments show a promising research result when using deep learning to predict learner performance in multi-question tasks and can ultimately provide more accurate adaptive tests for learners.