Purpose This paper aims to propose an AI/AL Bi-LSTM algorithm with Feature Weighting (FW) model to optimize educational resource recommendations. It addresses challenges like data sparsity, information loss and inefficient filtering in online learning platforms. The model integrates channel awareness and compressed interactive networks to refine data, assign feature weights and enhance relevance. Validated on Tencent Classroom, it achieves 98% categorization accuracy, 0.38 deficit function and 83% user satisfaction, outperforming baseline methods. The study highlights faster convergence, lower computational time and personalized recommendations based on user behavior. This approach improves e-learning efficiency and supports scalable, adaptive educational systems. Future work includes broader data set testing. Design/methodology/approach This study designs an AI/ML Bi-LSTM with Feature Weighting (FW) model for educational resource recommendations. The methodology involves: Data refinement through channel awareness to assign optimal feature weights, compressed interactive networks to reduce sparsity and enhance relevance and Bidirectional encoding via Bi-LSTM to minimize information loss. The model is trained on user interaction data (clicks, searches, video engagement) and validated against benchmarks (MLKR, FAMMC) using metrics like accuracy, deficit function and response time. Real-world testing on Tencent Classroom confirms its efficacy. The approach combines deep learning with feature optimization for scalable, personalized learning solutions. Findings The proposed Bi-LSTM with FW model demonstrates superior performance in educational resource recommendations. Key findings include: High Accuracy: Achieves 98% categorization accuracy on the Tencent Classroom data set. Efficiency: Maintains 0.38 deficit function and 350 ms response time, outperforming MLKR, FAMMC and Learner models. User satisfaction: Delivers 83% satisfaction through personalized recommendations. Robustness: Effectively handles sparse, high-dimensional data via feature weighting and Bi-LSTM encoding. Scalability: Shows stable performance across diverse educational resource types (videos, texts, courses). These results validate the model’s ability to reduce information loss, enhance relevance and adapt to user behavior, making it a scalable solution for e-learning platforms. Originality/value This study introduces three key innovations: Novel Architecture: First integration of Bi-LSTM with FW for educational resource recommendations, uniquely addressing data sparsity and information loss. Channel awareness mechanism: Original preprocessing technique that dynamically weights features to enhance relevance in learning data sets. Real-world validation: First AI model tested on Tencent Classroom demonstrating 98% accuracy while maintaining interpretability. The research advances personalized education technology by providing a computationally efficient alternative to traditional recommendation systems, introducing adaptable feature engineering for diverse learning resources and delivering actionable insights for ed-tech platforms through explainable. This work bridges the gap between theoretical deep learning and practical educational applications, offering measurable improvements over existing methods.
Ka’bi et al. (Thu,) studied this question.