Fine-Grained Classification of Localized Educational Resources Based on Deep Knowledge Tracing Model

Accurate representation and classification of sequential information are critical for intelligent engineering systems involving complex signal processing and adaptive decision-making. To address the data sparsity problem in fine-grained classification of localized educational resources, this study proposes a contrast-enhanced attentional knowledge tracing framework that combines Attentive Knowledge Tracing (AKT), sliding-window sequence augmentation, and InfoNCE-based contrastive learning. The proposed method extracts cognitive state transition vectors as latent representations of resource functions and performs hierarchical clustering to identify implicit instructional categories without relying on large-scale annotated corpora or textual features. Experimental results demonstrate that the proposed framework achieves a silhouette coefficient of 0.72 under full-data conditions and maintains robust clustering performance under extremely sparse data scenarios, significantly outperforming conventional knowledge tracing models. The learned cognitive representations exhibit weak correlation with coarse-grained subject labels while effectively capturing latent functional characteristics of educational resources. By providing an efficient mechanism for high-dimensional sequential representation learning and feature discrimination, the proposed framework offers valuable insights for educational data mining, adaptive instructional design, and data-driven resource classification strategies that are relevant to intelligent tutoring systems and personalized learning applications.