Metapath- and Entity-aware Graph Neural Network for Recommendation

Due to the shallow structure, classic graph neural networks (GNNs) fail in modelling high-order graph structures. Such high-order structures capture critical insights for downstream tasks. Concretely, in recommender systems, disregarding these insights lead to inadequate distillation of collaborative signals. In this paper, we employ collaborative subgraphs (CSGs) and metapaths to explicitly capture these high-order graph structures. We propose meta\textbf{P}ath and \textbf{E}ntity-\textbf{A}ware \textbf{G}raph \textbf{N}eural \textbf{N}etwork (PEAGNN). We extract an enclosing CSG for user-item pair within its $h$-hop neighbours. Multiple metapath-aware subgraphs are then extracted from CSG. PEAGNN trains multilayer GNNs to perform information aggregation on such subgraphs. This aggregated information from different metapaths is fused using attention mechanism. Finally, PEAGNN gives us the representations for node and subgraph, which can be used to train MLP for predicting score for target user-item pairs. To leverage the local structure of CSGs, we present entity-awareness that acts as a contrastive regularizer of node embedding. Moreover, PEAGNN can be combined with prominent layers such as GAT, GCN and GraphSage. Our empirical evaluation shows that our proposed technique outperforms competitive baselines on several datasets for recommendation task. Our analysis demonstrates that PEAGNN also learns meaningful metapath combinations from a given set of metapaths.