CORE: A causally-inspired framework for educational recommendation via contrastive learning

Online learning platforms face challenges of information overload and biased engagement, which hinder effective course recommendations. Traditional recommender systems struggle to distinguish genuine learner interests from confounding factors such as popularity, often leading to suboptimal suggestions. To address these issues, we propose a novel educational recommendation model that integrates causal contrastive learning. Our approach combines causal inference techniques with contrastive representation learning to disentangle a learner’s intrinsic interests from external biases. We formulate the task as learning user and course representations that remain invariant under causal interventions which remove popularity effects. The proposed model is evaluated on four benchmark datasets using Recall@10/20 and NDCG@10/20. Experimental results show consistent improvements of 2-3 percentage points over state-of-the-art methods on all datasets. In particular, our model outperforms strong baselines (including graph-based and contrastive models) on top-K recommendation accuracy. These gains demonstrate the effectiveness of integrating causal learning to enhance recommendation relevance and robustness. Our contributions include a new causal graph formulation for recommendations, a contrastive learning scheme based on counterfactual augmentations, extensive empirical validation against existing methods, and insightful analysis of how causal contrastive learning improves recommendation quality. The proposed framework offers a fresh perspective for developing next-generation recommender systems that are both accurate and causally aware, with significant implications for educational personalization.