Scholarlink

<p>This study presents a systematic literature review combined with unsupervised topic modeling and machine learning classification to map the intellectual landscape of intelligent adaptive e-learning systems. Following PRISMA guidelines, 67 peer-reviewed studies (2019–2025) were retrieved from five major databases and analyzed through a Python-based text mining pipeline incorporating custom lemmatization, domain-specific stopword filtering, and keyword deduplication. Grid search-optimized Latent Dirichlet Allocation (K = 2, α = 0.05, β = auto) yielded two latent topics: Explainable AI (29 articles, 43.3%) and Machine Learning and Cognitive Load (38 articles, 56.7%). A supervised binary classification model (SVM, AUC = 0.769) analyzed the distribution of seven critical research gaps across these thematic domains, with LIME interpretability analysis revealing domain-specific patterns. The results show that current literature converges around two complementary pillars: algorithmic transparency grounded in explainability principles, and cognitively informed instructional design integrating affective computing and multimodal data. HCI principles and real-time adaptation emerged as primary discriminating features between clusters, while five of seven gaps represent universal challenges shared across the field. Persistent gaps remain in authentic longitudinal evaluation, ethical governance of learner data, and standardized architectural frameworks. The study contributes methodologically by demonstrating the utility of LDA topic modeling combined with machine learning classification for augmenting systematic review rigor, and substantively by offering an evidence-based research agenda for developing scalable, pedagogically grounded, and learner-centered adaptive e-learning architectures.</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

A systematic review of intelligent adaptive e-learning systems using LDA topic modeling and machine learning classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月19日

查看全文
  • Agostino Marengo,
  • Douglas Sales Alves Amante,
  • Alessandro Pagano,
  • Miltiades Demetrios Lytras,
  • Jenny Pange,
  • Vito Santamato

摘要

This study presents a systematic literature review combined with unsupervised topic modeling and machine learning classification to map the intellectual landscape of intelligent adaptive e-learning systems. Following PRISMA guidelines, 67 peer-reviewed studies (2019–2025) were retrieved from five major databases and analyzed through a Python-based text mining pipeline incorporating custom lemmatization, domain-specific stopword filtering, and keyword deduplication. Grid search-optimized Latent Dirichlet Allocation (K = 2, α = 0.05, β = auto) yielded two latent topics: Explainable AI (29 articles, 43.3%) and Machine Learning and Cognitive Load (38 articles, 56.7%). A supervised binary classification model (SVM, AUC = 0.769) analyzed the distribution of seven critical research gaps across these thematic domains, with LIME interpretability analysis revealing domain-specific patterns. The results show that current literature converges around two complementary pillars: algorithmic transparency grounded in explainability principles, and cognitively informed instructional design integrating affective computing and multimodal data. HCI principles and real-time adaptation emerged as primary discriminating features between clusters, while five of seven gaps represent universal challenges shared across the field. Persistent gaps remain in authentic longitudinal evaluation, ethical governance of learner data, and standardized architectural frameworks. The study contributes methodologically by demonstrating the utility of LDA topic modeling combined with machine learning classification for augmenting systematic review rigor, and substantively by offering an evidence-based research agenda for developing scalable, pedagogically grounded, and learner-centered adaptive e-learning architectures.