Temporal dynamics of user preferences in learning content recommendation systems using SCL1-LSTM technique

This study proposes a personalized Content Recommendation System (CRS) that adapts to users’ evolving interests and learning needs. It introduces the Saturated Capped L1-Long Short-Term Memory (SCL1-LSTM) classifier to improve recommendation accuracy. The system integrates CRS database content with Twitter review data to capture both structured and unstructured insights. K-Laplacian Polynomial Clustering (KLaPMC) is employed to group learning content by difficulty and user preferences. Apache Spark structures the grouped data, which is then collaboratively filtered and mapped. Singular Value Decomposition (SVD) and Regularized Quadratic Correlation (RQC) are applied to uncover linear and nonlinear latent user-item features. Feature extraction and time deviation analysis follow, with the Globally Guided Golden Eagle Optimizer (3GEO) selecting optimal features to enhance performance. The SCL1-LSTM model, incorporating sparsity and advanced activation functions, achieves a recommendation accuracy of 97.63%, outperforming conventional methods. This CRS framework effectively personalizes learning content by considering both temporal dynamics and user feedback.