Recent advances in Learning Analytics have shown the importance of modeling complex relationships among learners, learning resources, and interactions in online education environments. Graph-based representations naturally capture these multi-relational structures, enabling more accurate analysis of learning behaviors and personalized support. However, existing graph neural network models face challenges in terms of computational scalability and interpretability, which limit their applicability in large-scale e-learning platforms. To address these issues, this paper introduces two novel kernel-based methods for graph representation learning: the Simplified Graph Neural Tangent Kernel (SGTK) and the Simplified Graph Neural Kernel (SGNK). SGTK replaces the traditional multi-layer stacking mechanism with a continuous K-step aggregation procedure, reducing redundant computations while maintaining expressive power. SGNK further advances this approach by modeling infinitely wide graph neural networks as Gaussian Processes, allowing kernel values to be computed directly without explicit layer-wise operations. Applied to learning analytics tasks such as student performance prediction and resource recommendation, our methods demonstrate competitive accuracy compared to state-of-the-art graph neural models, while achieving superior computational efficiency. By integrating graph kernel methods with learning analytics, this work offers an efficient approach for analyzing complex learner-resource-interaction networks in e-learning environments. Extensive experiments on node classification task demonstrate that the proposed SGTK and SGNK achieve performance comparable to existing approaches while improving computational efficiency. Implementation details are available at https://github.com/WANGLin0126/SGNK .