Scholarlink

<p>This study introduces an explainable artificial intelligence (XAI) framework that enhances the transparency of educational question classification based on Bloom’s taxonomy. The proposed system integrates a convolutional neural network (CNN) with SHAP and LIME to provide both global and local explanations for each prediction. The model was trained on a balanced dataset of 5000 manually labeled questions and achieved strong performance, including 88% accuracy, 87% precision, 86% recall, and an 86% F1-score. Beyond predictive performance, the integration of SHAP and LIME proved essential for revealing the reasoning behind the CNN’s decisions. SHAP consistently highlighted action verbs and conceptual cues aligned with Bloom’s cognitive levels, while LIME provided instance-level explanations that clarified the influence of specific words on each prediction. Expert evaluation confirmed that these explanations were pedagogically valid, interpretable, and aligned with human reasoning. The findings demonstrate that the incorporation of XAI significantly increases model trustworthiness and supports its practical adoption in assessment design and instructional decision-making.</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

Enhancing transparency in educational question classification: an explainable AI approach using SHAP and LIME with CNN models

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Dima R. Badawi

摘要

This study introduces an explainable artificial intelligence (XAI) framework that enhances the transparency of educational question classification based on Bloom’s taxonomy. The proposed system integrates a convolutional neural network (CNN) with SHAP and LIME to provide both global and local explanations for each prediction. The model was trained on a balanced dataset of 5000 manually labeled questions and achieved strong performance, including 88% accuracy, 87% precision, 86% recall, and an 86% F1-score. Beyond predictive performance, the integration of SHAP and LIME proved essential for revealing the reasoning behind the CNN’s decisions. SHAP consistently highlighted action verbs and conceptual cues aligned with Bloom’s cognitive levels, while LIME provided instance-level explanations that clarified the influence of specific words on each prediction. Expert evaluation confirmed that these explanations were pedagogically valid, interpretable, and aligned with human reasoning. The findings demonstrate that the incorporation of XAI significantly increases model trustworthiness and supports its practical adoption in assessment design and instructional decision-making.