Scholarlink

<p>As educational environments become increasingly heterogeneous, conventional teaching strategies often fall short in accommodating the diverse and evolving learning behaviors of students, particularly when individual learning preferences are ambiguous or not explicitly expressed. To address this growing complexity, we introduce SRE-TransformerNet, a robust AI-driven framework designed to foster personalized and inclusive educational experiences. This framework synergistically integrates the capabilities of Swin Transformer, ResNet, and EfficientNet, enabling it to dynamically identify and respond to variations in learning styles. To further optimize data processing, three novel preprocessing techniques are embedded into the pipeline. The Adaptive Range Scaling (ARS) method ensures consistency in feature distribution. Feature Fusion and Weight Adjustment (FFWA) enhances the relevance of selected features, while Uncertainty-Driven Transformation (UDT) is specifically designed to improve model resilience in scenarios involving incomplete or ambiguous data. To comprehensively assess the effectiveness of the model, we introduce three new performance metrics. The Categorical Similarity Score (CSS) evaluates inter-class pattern recognition, the Temporal Consistency Index (TCI) captures the model’s stability over time, and the Multi-Class Imbalance Metric (MCIM) quantifies the system’s robustness in imbalanced learning scenarios. Empirical evaluations demonstrate that SRE-TransformerNet achieves high predictive performance, with an F1-score of 0.987, AUC of 0.995, and accuracy of 0.988, alongside a recall rate of 0.986. These results underscore the model’s efficacy in minimizing misclassifications across diverse student profiles. Ultimately, the framework has strong potential to support adaptive learning systems−whether deployed in intelligent tutoring environments or large-scale e-learning platforms−offering a scalable solution for enhancing educational equity and effectiveness in real time.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Transformer-based deep learning for adaptive pedagogy under uncertain student preferences

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月26日

查看全文
  • Huan Wang

摘要

As educational environments become increasingly heterogeneous, conventional teaching strategies often fall short in accommodating the diverse and evolving learning behaviors of students, particularly when individual learning preferences are ambiguous or not explicitly expressed. To address this growing complexity, we introduce SRE-TransformerNet, a robust AI-driven framework designed to foster personalized and inclusive educational experiences. This framework synergistically integrates the capabilities of Swin Transformer, ResNet, and EfficientNet, enabling it to dynamically identify and respond to variations in learning styles. To further optimize data processing, three novel preprocessing techniques are embedded into the pipeline. The Adaptive Range Scaling (ARS) method ensures consistency in feature distribution. Feature Fusion and Weight Adjustment (FFWA) enhances the relevance of selected features, while Uncertainty-Driven Transformation (UDT) is specifically designed to improve model resilience in scenarios involving incomplete or ambiguous data. To comprehensively assess the effectiveness of the model, we introduce three new performance metrics. The Categorical Similarity Score (CSS) evaluates inter-class pattern recognition, the Temporal Consistency Index (TCI) captures the model’s stability over time, and the Multi-Class Imbalance Metric (MCIM) quantifies the system’s robustness in imbalanced learning scenarios. Empirical evaluations demonstrate that SRE-TransformerNet achieves high predictive performance, with an F1-score of 0.987, AUC of 0.995, and accuracy of 0.988, alongside a recall rate of 0.986. These results underscore the model’s efficacy in minimizing misclassifications across diverse student profiles. Ultimately, the framework has strong potential to support adaptive learning systems−whether deployed in intelligent tutoring environments or large-scale e-learning platforms−offering a scalable solution for enhancing educational equity and effectiveness in real time.