Cognitive diagnosis infers students' mastery of concepts from response logs, yet performance is also shaped by non‑cognitive factors such as emotion, engagement, and fatigue. Affective cognitive diagnosis extends the conventional framework by incorporating affective states. Existing approaches often assume that the cognitive backbone already accounts for student, item, and concept effects, attributing the remaining error mainly to affect. We argue that real educational data contain stable cognitive residuals caused by item calibration bias, systematic concept bias, personalized student‑concept deviations, and latent student‑item matching. Without an explicit pathway, these residuals can leak into affective representations, resulting in affect contamination. To solve this, we propose an ability‑residual decoupled framework. The model first captures unmodeled cognitive residuals via student, item, concept, student‑concept, and low‑rank student‑item components, then an affective module modulates guess and slip effects. A Q‑matrix‑constrained concept residual attention mechanism adaptively aggregates only the concept residuals relevant to the current item. Experiments on ASSIST2017, ASSIST2012, ASSIST2009, and Junyi, using six cognitive diagnosis backbones, consistently improve response prediction and, when affect labels are available, achieve better affect alignment. Ablation studies, leakage probes, PCA visualizations, long‑tail analysis, and case studies demonstrate that ability residuals absorb stable cognitive bias, reduce cognitive contamination in the affective branch, and enhance robustness and predictive accuracy.
Review