Heterogeneous Graph Prototypical Networks for Few-Shot Node Classification

The node classification task is one of the most significant applications in heterogeneous graph analysis, which is widely used for modeling multi-typed interactions. Meanwhile, Graph Neural Networks (GNNs) have aroused wide interest due to their remarkable effects on graph node classification. However, there are some challenges when applying GNNs to heterogeneous graph node classification: the cumbersome node labeling cost, and the heterogeneity of graphs. Existing GNNs require sufficient annotation while learning classifiers independently with node embeddings cannot exploit graph topology effectively. Recently, few-shot learning has achieved competitive results in homogeneous graphs to address the performance degradation in the label sparsity case. While heterogeneous graph few-shot learning is limited by the difficulties of extracting multiple semantics. To this end, we propose a novel Heterogeneous graph Prototypical Network (HPN) with two modules: Graph structural module generates node embeddings and semantics for meta-training by capturing heterogeneous structures. Meta-learning module produces prototypes with heterogeneous induced subgraphs for meta-training classes, which improves knowledge utilization compared with the traditional meta-learning. Experimental results on three real-world heterogeneous graphs demonstrate that HPN achieves outstanding performance and better stability.