Multi-Sampling Item Response Ranking Neural Cognitive Diagnosis with Bilinear Feature Interaction

Cognitive diagnosis is a fundamental task in educational data mining that aims to discover students' proficiency in knowledge concepts. Neural cognitive diagnosis combines deep learning with cognitive diagnosis, breaking away from artificially defined interaction functions. However, existing cognitive diagnosis models mostly start from the interaction of students' answers, ignoring the feature interaction between test items and knowledge concepts. Meanwhile, few of the previous models consider the monotonicity of knowledge concept proficiency. To address these issues, we present a novel cognitive diagnosis method, called multi-sampling item response ranking neural cognitive diagnosis with bilinear feature interaction. We first allow the ratio in loss function to adjust the impact between pointwise sampling and pairwise sampling to strengthen the monotonicity. At the same time, we replace element product feature interaction with bilinear feature interaction in the multi-sampling item response ranking neural cognitive diagnosis to enhance interaction in the deep learning process. Specifically, our model is stable and can be easily applied to cognitive diagnosis. We observed improvements over the previous state-of-the-art baselines on real-world datasets.