Fine-Grained Intelligent Learning Diagnosis Model Based on the Exercise–Knowledge–Cognition Tensor for Educational Assessment

Accurate and interpretable learning diagnosis is increasingly required in AI-enabled educational assessment. Existing cognitive diagnostic models typically represent item attributes with a binary Q-matrix and infer mastered or not mastered knowledge states. Although polytomous extensions allow graded mastery, item attributes rarely encode theory-aligned cognitive-process demands, which limits pedagogical interpretation of diagnosed profiles. This study aims to operationalize revised Bloom’s taxonomy at the exercise–knowledge level by constructing an Exercise–Knowledge–Cognition tensor and to develop RLDM-EKC as a DINA-type cognitive diagnosis model that infers ordered knowledge–cognition profiles. The model defines EKC-based ideal responses, estimates slip and guess parameters with an Expectation–Maximization procedure, and derives learner profiles using Maximum A Posteriori inference with uncertainty summaries. We validate the approach on synthetic data and on TIMSS 2007 Grade 4 mathematics data, comparing against classical CDMs including DINA, PA-DINA, and pG-DINA. In simulation, RLDM-EKC attains a PMR of 81.7% and an AAMR of 91.6%, and in empirical data, it yields theory-aligned multi-level cognitive profiles with transparent uncertainty reporting. These properties support actionable, human-in-the-loop feedback for teachers and learners under realistic deployment constraints.