With the introduction of online learning systems into education, the use of accumulated educational big data has become a critical issue in recent years. In the field of artificial intelligence, adaptive learning, which provides appropriate learning support for each learner, has attracted attention. Applying machine learning methods to educational big data to predict learners' responses to task items, thereby understanding learners' strengths and weaknesses, can contribute to active learning. The authors' previously proposed DeepIRT combines deep learning methods with Item Response Theory (IRT) to estimate parameters that represent learners' abilities and item difficulties using separate neural networks, achieving both high accuracy in predicting responses to unknown items and high interpretability.

This paper proposes a method that enhances prediction accuracy by combining DeepIRT with the deep learning method Hypernetwork, commonly used in the field of natural language processing. The introduction of Hypernetwork updates parameters for attenuating the feature vectors extracted from past training data when estimating the learners' abilities, balancing the importance of past training data with the latest training data. The prediction accuracy of the proposed method is validated through experiments with real-world data, showing significant improvement over previous methods, especially in long-term learning processes. Furthermore, experiments with simulated data demonstrate that the learners' abilities estimated by the proposed method strongly correlate with actual ability parameters compared to other methods, indicating superior interpretability.

This paper is highly evaluated for achieving both high prediction accuracy that surpasses state-of-the-art methods and interpretability in response prediction estimation with the carefully designed method. The idea of updating some of the parameters used in the existing method using Hypernetwork shows exceptional originality. Based on the proposed method, appropriate educational interventions in adaptive learning should become possible, marking an important contribution to the field of educational technology.