<p>The application of deep learning in education enables intelligent recommendation and precise matching of resources to learner needs, promoting personalized and adaptive learning. With the rapid growth of digital content, ensuring the delivery of relevant and accurate resources has become a critical research challenge. Existing recommendation methods often suffer from data sparsity, limited semantic understanding, and limited adaptability to diverse learner contexts. These limitations reduce the precision and effectiveness of resource allocation. To address these issues, this paper proposes a Knowledge Graph-Enhanced Recommender CNN (KG-RecCNN) framework that integrates semantic resource mapping via knowledge graphs with deep CNN embeddings for precise feature representation and relevance matching. This hybrid approach captures both explicit relationships and latent learning patterns. The proposed method is applied to educational platforms to recommend courses, materials, and assessments tailored to individual learner profiles. It enables dynamic adaptation by aligning resource semantics with learner preferences and knowledge levels. Experimental findings demonstrate that KG-RecCNN significantly improves recommendation accuracy, relevance, and learner satisfaction compared to traditional approaches. The results highlight its effectiveness in advancing intelligent education systems through precise, context-aware resource matching. KG-RecCNN achieved 88.3% accuracy, 87.9% F1-score, 0.85 semantic relevance, 0.82 adaptability, 4.6/5 learner satisfaction, and 35% faster computation, outperforming RMM, DNRM, and MLPEF significantly.</p>