CogniSkillNet: A Multi-Domain Cognitive Skill-Task Network for Criterion-Referenced Assessment

The assessment of cognitive skills across multipledomains remains a significant challenge in educational measurement and intelligent tutoring systems. Traditional cognitivediagnostic models often struggle with capturing complex nonlinear interactions between learners and tasks while maintaining interpretability. We propose CogniSkillNet, a novel neuralnetwork-based framework that integrates multi-domain cognitiveskill assessment with criterion-referenced evaluation principles.Our approach leverages deep learning architectures to modelintricate skill-task relationships while incorporating educationalpriors and monotonicity constraints to ensure interpretability.The framework employs a hierarchical attention mechanism tocapture both local skill mastery and global cognitive competencies across diverse domains. Through extensive experiments onreal-world educational datasets spanning mathematics, languagearts, and scientific reasoning, CogniSkillNet achieves superiorperformance with 7.3% improvement in diagnostic accuracycompared to state-of-the-art methods while providing transparent, criterion-referenced feedback for personalized learninginterventions.