<p>Open Educational Resources (OER) platforms are confronted with the challenges of information overload, data sparsity, and heterogeneous learner’s preferences, which hinder the achievement of accurate and adaptive recommendation. Existing methodologies or algorithms are inadequate to capture these behaviors dynamically, resulting in relatively poor accuracies and low responsivity. To solve this issue, an AI-driven personalized recommendation approach is proposed, which improves the relevance, adaptability, and runtime performance in open education environments. The goal is to design an efficient pipeline to derive a diverse set of learner and resource meta-data from sources such as videos, textual modules, and shared materials. The acquired dataset is pre-processed for noise removal, normalization and missing values and outliers are treated. As a form of textual representation, Term Frequency–Inverse Document Frequency (TF-IDF) vectorization is used to represent the content as a weighted feature vector by giving more weight to important words and less to the common background’s words. The extracted features are then used to the proposed Egret Swarm Optimization (ESO) based Dynamic Bidirectional Long Short-Term Memory (DBi-LSTM) model, wherein ESO optimizes the hyperparameters for achieving faster convergence and the DBi-LSTM model captures the sequential learning process through forward and backward temporal flows. So, this allows the system to have much more accurate and context-aware recommendations. Simulation results confirm an achieved accuracy of 93.6%, a precision of 92.5%, a recall of 91%, and an F1-score of 90.2%, outperforming many traditional methods, thus substantiating the efficacy of the proposed AI-based framework for personalized OER recommendation.</p>