Scholarlink

The fast-paced growth of online learning platforms necessitates the existence of potent recommendation systems to help guide learners through the vast oceanic catalogs of courses, with a resultant enhancement of engagement and learning outcomes. However, the selection of e-learning content is challenged by data sparsity and the cold-start problem. This comprehensive work offers a comparative analysis of the eighteen recommendation models with the widest application in an e-learning setting against a merged dataset of Coursera and Udemy course information. The recommendation models are grouped into three main approaches: Content-Based Filtering, Collaborative Filtering, and Hybrid methods, reflecting both classical Machine Learning (ML) techniques as well as recent Deep Learning (DL) ones in every single category (six models in total per category, evenly distributed among the two approaches, ML and DL). Model performance evaluation is based on RMSE, MAE, Precision@5, Recall@5, and other metrics to capture various aspects of model effectiveness. In addition, model interpretability has been enhanced by applying SHAP (SHapley Additive exPlanations) analysis to gain insights into the contributions of features toward the recommendations. Findings show that models, per type and metric, produced different results indicative of the benefits associated with hybrid ones and the strength of DL models, with SHAP analysis providing meaningful insights on how recommendations are generated. This study provides a broad empirical comparison to serve, among others, as a guide in selecting and developing robust and explainable e-learning recommendation systems.

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

Comparative Analysis of Machine Learning and Deep Learning Models for E-Learning Recommendation Systems with SHAP Explainability

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Bhanushray Gupta,
  • G. Gnanasabesan,
  • T. V. S. Praneeth,
  • Sajja Naga Koushik,
  • T. Keerthika

摘要

The fast-paced growth of online learning platforms necessitates the existence of potent recommendation systems to help guide learners through the vast oceanic catalogs of courses, with a resultant enhancement of engagement and learning outcomes. However, the selection of e-learning content is challenged by data sparsity and the cold-start problem. This comprehensive work offers a comparative analysis of the eighteen recommendation models with the widest application in an e-learning setting against a merged dataset of Coursera and Udemy course information. The recommendation models are grouped into three main approaches: Content-Based Filtering, Collaborative Filtering, and Hybrid methods, reflecting both classical Machine Learning (ML) techniques as well as recent Deep Learning (DL) ones in every single category (six models in total per category, evenly distributed among the two approaches, ML and DL). Model performance evaluation is based on RMSE, MAE, Precision@5, Recall@5, and other metrics to capture various aspects of model effectiveness. In addition, model interpretability has been enhanced by applying SHAP (SHapley Additive exPlanations) analysis to gain insights into the contributions of features toward the recommendations. Findings show that models, per type and metric, produced different results indicative of the benefits associated with hybrid ones and the strength of DL models, with SHAP analysis providing meaningful insights on how recommendations are generated. This study provides a broad empirical comparison to serve, among others, as a guide in selecting and developing robust and explainable e-learning recommendation systems.