Answers are grounded in this paper and cite their sources.
With the rapid growth of educational data, analysing students’ learning pathways has become increasingly challenging. Traditional learning behaviour analysis methods struggle to handle multi-channel and heterogeneous data, and they often fail to capture complex relationships among student behaviours, particularly in multi-dimensional feature fusion and spatiotemporal dependency modelling. To address these limitations, this paper proposes PathNet-GC, a learning path prediction model that integrates multi-channel temporal feature extraction with graph convolutional networks (GCNs). Specifically, the model first extracts temporal representations from multi-channel learning behaviour data using 1D convolution and Bi-GRU, and then employs a GCN module to model structured relationships among students’ behaviours. Through the deep fusion of temporal features and graph-based behavioural dependencies, PathNet-GC improves the accuracy, stability, and robustness of learning path prediction. Experiments conducted on the EdNet and ASSISTments datasets demonstrate that PathNet-GC outperforms traditional machine learning methods and representative deep learning baselines, especially in terms of F1-score and AUC. Overall, this study not only enhances the effectiveness of learning pathway analysis, but also provides a feasible modelling framework and practical methodological reference for educational data mining, personalized learning support, and intelligent education systems in diverse real-world educational scenarios.
No takes yet. Share an insight, caveat, or question.
Gan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: