Abstract

With the growth of e-learning content at a disproportionate rate, the need for recommendation systems that personalize content delivery and address information overload has become critical. This work explores information technology integration in providing intelligent recommendation systems for e-learning contents using machine learning as an architecture in the science of design and experiment as an evaluation research. A hybrid recommendation architecture was defined to establish Neural Collaborative Filtering, BERT content embeddings, and LSTM sequential modeling combined with an evaluation of multiple baseline algorithms: user-based collaborative filtering, item-based collaborative filtering, matrix factorization, and content-based system. The hybrid architecture was trained and evaluated using a large benchmark dataset of 5,000 learners and 2,000 learning objects; accuracy, diversity, and user-centered evaluations were used to evaluate the recommendation accuracy (Precision@K, Recall@K, NDCG@K), diversity, and user-centered evaluation performance, respectively. The results showed that the hybrid approach achieved the highest accuracy with Precision@10 of 0.847, Recall@10 of 0.792, and NDCG@10 of 0.863, significantly outperforming all the base algorithms (p <.001). User-centered evaluation indicated a 47.3% click-through rate increase, a 31.8% completion rate increase, and a 38.7% session duration increase over popularity. The IT infrastructure such as cloud computing, Apache Spark, MongoDB, and Redis optimized scalable real-time recommendation delivery with an average inference latency of 42ms. Hard problems such as cold-start in 8.3% of users and 14.7% of learning objects were alleviated by content-based fallback and cross-domain transfer learning approaches. This work demonstrates that technology, when built within an architecture that emphasizes diversity and serendipity, can substantially improve e-learning content design. Originality is in the design science research study, cross-domain transfer learning, and impact studies using experiment research.