Recommendation System Model for E-Learning Using Hybrid Deep Learning Models

Abstract

The use of personalized learning resource recommendation is a vital aspect of boosting learner engagement which helps in effective e-learning. But current recommendation systems come with a number of problems such as information overload, modelling learner preferences, and making effective use of learner feedback. This research aims to propose a hybrid recommendation system for learning resources in e-learning based on learner profiling, Natural Language Processing (NLP), collaborative filtering and intelligent assistant. The framework is used to analyse the learner characteristics, preferences, interaction pattern and textual feedback, identifying learning resources and recommending the appropriate ones. Confusion-matrix analysis, ROC analysis and ranking-oriented recommendation measures were used for experimental evaluation. The proposed framework obtained an accuracy of 57% and a precision of 52.83%, a recall of 60.87%, and an F1-score of 56.57%. The ROC analysis yielded an AUC of 0.57 and the statistical validation gave a 95% Wilson confidence interval of 47.2%–66.3% for the observed accuracy. Furthermore, a comparison of the framework with benchmark approaches reported in the literature was conducted. The performance of the benchmark approaches (EELR and DELRNet) was higher in the respective experimental setups, but these comparisons were made in a context, as the benchmark models were not re-implemented in the same datasets, preprocessing, partitions, and experiments. The results show that it is feasible to combine these four technology components—learner profiling, NLP, collaborative filtering, and intelligent assistance—into a single e-learning recommendation system, and that further expansion of the data set, with controlled experiments set to benchmark the results, and optimization of the system will be needed to enhance the quality of the recommendations and their scalability.