Scholarlink

<p>The rapid integration of Generative Artificial Intelligence (GenAI) into education necessitates advanced stochastic frameworks capable of modeling the complex and probabilistic nature of learning trajectories. This paper introduces a <i>Semi-Markov Hidden Markov Model</i> (HMM-S) that explicitly parameterizes state durations using Gamma distributions and integrates conversational covariates such as explanation richness, feedback immediacy, and conceptual density. Leveraging heterogeneous datasets from Khan Academy, StudyChat, OULAD, and ASSISTments across both K-12 and higher education contexts, the proposed model captures the multimodal interactions between learners and GenAI tutors. Empirical analyses reveal that explanation richness and immediate feedback jointly enhance confusion-to-comprehension transitions by up to 41%, demonstrating the superiority of the HMM-S over classical and deep knowledge tracing baselines. The study advances stochastic modeling of educational processes, contributes to equitable GenAI design through ethical transparency mechanisms, and establishes a quantitative foundation for adaptive, data-driven pedagogy in global learning ecosystems. The framework provides educators with actionable insights for implementing adaptive GenAI tutoring systems that balance pedagogical effectiveness with ethical considerations across diverse global contexts.</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

Modeling learning trajectories in GenAI-augmented education: a semi-Markov hidden Markov framework for ethical and adaptive pedagogy

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • Mohamed Yasser Bounnite

摘要

The rapid integration of Generative Artificial Intelligence (GenAI) into education necessitates advanced stochastic frameworks capable of modeling the complex and probabilistic nature of learning trajectories. This paper introduces a Semi-Markov Hidden Markov Model (HMM-S) that explicitly parameterizes state durations using Gamma distributions and integrates conversational covariates such as explanation richness, feedback immediacy, and conceptual density. Leveraging heterogeneous datasets from Khan Academy, StudyChat, OULAD, and ASSISTments across both K-12 and higher education contexts, the proposed model captures the multimodal interactions between learners and GenAI tutors. Empirical analyses reveal that explanation richness and immediate feedback jointly enhance confusion-to-comprehension transitions by up to 41%, demonstrating the superiority of the HMM-S over classical and deep knowledge tracing baselines. The study advances stochastic modeling of educational processes, contributes to equitable GenAI design through ethical transparency mechanisms, and establishes a quantitative foundation for adaptive, data-driven pedagogy in global learning ecosystems. The framework provides educators with actionable insights for implementing adaptive GenAI tutoring systems that balance pedagogical effectiveness with ethical considerations across diverse global contexts.