Aspect-Aware MOOC Recommendation in a Heterogeneous Network

TL;DR

AMR addresses the limitations of hand-crafted MOOC metapaths by automatically discovering metapaths via bi-directional walks and learning path-specific aspect representations with a bi-LSTM. It then aggregates these representations as edge features in a learner-learner KC-KC subgraph and uses a GCN backbone to predict learner-KC ratings with a joint loss combining triplet and BPR signals. Experiments on MOOCCube and PEEK show AMR consistently outperforms state-of-the-art GNN baselines across HR@K and nDCG@K, and analyses reveal richer, path-specific semantics across aspects. The approach offers scalable, semantically informed KC recommendations that better align with learners' conceptual needs and demonstrates the value of aspect-aware heterogeneous graph modeling for education.

Abstract

MOOC recommendation systems have received increasing attention to help learners navigate and select preferred learning content. Traditional methods such as collaborative filtering and content-based filtering suffer from data sparsity and over-specialization. To alleviate these limitations, graph-based approaches have been proposed; however, they still rely heavily on manually predefined metapaths, which often capture only superficial structural relationships and impose substantial burdens on domain experts as well as significant engineering costs. To overcome these limitations, we propose AMR (Aspect-aware MOOC Recommendation), a novel framework that models path-specific multiple aspects by embedding the semantic content of nodes within each metapath. AMR automatically discovers metapaths through bi-directional walks, derives aspect-aware path representations using a bi-LSTM-based encoder, and incorporates these representations as edge features in the learner-learner and KC-KC subgraphs to achieve fine-grained semantically informed KC recommendations. Extensive experiments on the large-scale MOOCCube and PEEK datasets show that AMR consistently outperforms state-of-the-art graph neural network baselines across key metrics such as HR@K and nDCG@K. Further analysis confirms that AMR effectively captures rich path-specific aspect information, allowing more accurate recommendations than those methods that rely solely on predefined metapaths. The code will be available upon accepted.

Figures (5)

• Figure 1: Overall Architecture of AMR. The framework consists of four key components which are (1) path generation module, (2) aspect-based representation module, (3) aspect aggregator, and (4) aspect importance estimation module.
• Figure 2: Performance changes on MOOCCube measured by different metrics including HR@K and nDCG@K depending on the number of aspects. The performance improves as the number of aspects increases showing the best performance with eight aspects.
• Figure 3: Performance changes on MOOCCube measured by different metrics including HR@K and nDCG@K depending on the path length. Model performance remains unchanged as path length varies.
• Figure 4: Distribution of aspect importance for learner and KC representations. KC representations show evenly distributed aspect importance, whereas learner representations are dominated by a few key aspects.
• Figure 5: Diversity of edge features within identical metapaths for learners and KCs. Feature embeddings are 50-dimensional, with colors indicating feature values. The embeddings vary markedly across both learner and KC pairs.