Scholarlink

<p>Current automated evaluation systems in virtual learning environments often operate as isolated modules with limited semantic understanding, no explanatory capabilities, and poor integration with adaptive tutoring mechanisms. These limitations hinder their ability to provide actionable and interpretable feedback to students within real-world educational platforms, such as Moodle. In response to this gap, we developed and validated an explainable natural language processing (NLP)-based educational assistant that integrates semantic analysis, justification of evaluations, and adaptive tutoring into a unified, modular architecture embedded in Moodle. The system utilizes transformer-based models (BERT and CodeBERT) to analyze academic submissions, generates traceable feedback using explainable AI techniques (LIME and SHAP), and activates personalized tutoring actions based on a dynamic adaptive index (A<sub>i</sub>). Experimental validation was conducted with 90 students, using real academic submissions and rubrics. Results show a mean absolute error of 0.34 and a Cohen’s Kappa of 0.78 in semantic evaluation. The system achieved a semantic explanation coverage index (C<sub>e</sub>) of 0.85, with 91% of tutoring activations deemed appropriate for explanation. The mean system response time was 2.78&#xa0;s across all modules, with 93.4% of actions executed within predefined technical thresholds. The global integrity index <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="44163_2025_438_Article_IEq1.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="13" /> </InlineMediaObject> <EquationSource Format="TEX">\(\Omega \)</EquationSource> </InlineEquation>, synthesizing performance across all modules, reached 0.806. This work demonstrates that it is technically feasible to implement a fully explainable, adaptive, and efficient educational assistant for real-time operation within institutional Learning Management Systems (LMS). The proposed architecture advances the integration of NLP and XAI in educational automation by combining semantic precision, traceability, and adaptive responsiveness in a single validated system.</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

Explainable educational assistant integrated in Moodle: automated semantic assessment and adaptive tutoring based on NLP and XAI

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月30日

查看全文
  • William Villegas-Ch,
  • Rommel Gutierrez,
  • Joselin García-Ortiz,
  • Vanessa Guevara

摘要

Current automated evaluation systems in virtual learning environments often operate as isolated modules with limited semantic understanding, no explanatory capabilities, and poor integration with adaptive tutoring mechanisms. These limitations hinder their ability to provide actionable and interpretable feedback to students within real-world educational platforms, such as Moodle. In response to this gap, we developed and validated an explainable natural language processing (NLP)-based educational assistant that integrates semantic analysis, justification of evaluations, and adaptive tutoring into a unified, modular architecture embedded in Moodle. The system utilizes transformer-based models (BERT and CodeBERT) to analyze academic submissions, generates traceable feedback using explainable AI techniques (LIME and SHAP), and activates personalized tutoring actions based on a dynamic adaptive index (Ai). Experimental validation was conducted with 90 students, using real academic submissions and rubrics. Results show a mean absolute error of 0.34 and a Cohen’s Kappa of 0.78 in semantic evaluation. The system achieved a semantic explanation coverage index (Ce) of 0.85, with 91% of tutoring activations deemed appropriate for explanation. The mean system response time was 2.78 s across all modules, with 93.4% of actions executed within predefined technical thresholds. The global integrity index \(\Omega \) , synthesizing performance across all modules, reached 0.806. This work demonstrates that it is technically feasible to implement a fully explainable, adaptive, and efficient educational assistant for real-time operation within institutional Learning Management Systems (LMS). The proposed architecture advances the integration of NLP and XAI in educational automation by combining semantic precision, traceability, and adaptive responsiveness in a single validated system.