Using Subgraph GNNs for Node Classification:an Overlooked Potential Approach

TL;DR

This work tackles the scalability limitations of global GNNs by reframing node classification as subgraph classification through SubGND, which integrates differentiated zero-padding, an ego-alter subgraph representation, and an adaptive feature scaling mechanism. By generating informative induced subgraphs via ISRW and fusing ego and alter information with a GIN backbone, SubGND achieves competitive performance with global GNNs, particularly excelling in heterophilic graphs where traditional methods struggle. Comprehensive experiments on six benchmarks show SubGND matching or surpassing state-of-the-art models, with notable gains on heterophilic datasets like Squirrel and Chameleon, and ablations confirm the importance of each component for robustness. The approach promises scalable, high-accuracy node classification across varied graph structures without extensive architecture-specific adjustments.

Abstract

Previous studies have demonstrated the strong performance of Graph Neural Networks (GNNs) in node classification. However, most existing GNNs adopt a node-centric perspective and rely on global message passing, leading to high computational and memory costs that hinder scalability. To mitigate these challenges, subgraph-based methods have been introduced, leveraging local subgraphs as approximations of full computational trees. While this approach improves efficiency, it often suffers from performance degradation due to the loss of global contextual information, limiting its effectiveness compared to global GNNs. To address this trade-off between scalability and classification accuracy, we reformulate the node classification task as a subgraph classification problem and propose SubGND (Subgraph GNN for NoDe). This framework introduces a differentiated zero-padding strategy and an Ego-Alter subgraph representation method to resolve label conflicts while incorporating an Adaptive Feature Scaling Mechanism to dynamically adjust feature contributions based on dataset-specific dependencies. Experimental results on six benchmark datasets demonstrate that SubGND achieves performance comparable to or surpassing global message-passing GNNs, particularly in heterophilic settings, highlighting its effectiveness and scalability as a promising solution for node classification.

Figures (3)

• Figure 1: Overall Framework: a) Induced subgraphs are constructed using a slightly modified variant of the ISRW algorithm, with each subgraph labeled by the ego-vertex's label. b) Node representations are augmented by distinct zero-padding operations are applied separately to the ego-vertex and alter-vertex. c) Finally, node embeddings are updated via a GNN encoder, and the ego-vertex embedding is concatenated with pooled representations of alter-vertex set for the final prediction.
• Figure 2: (a) Different types of ego-vertex may possess identical neighboring subgraphs, leading to the introduction of conflicting information into the model. (b) Through the application of Random Walk with Restart (RWR) and decomposed representations, We can gain more differentiated perspectives to alleviate this problem
• Figure 3: Parallel Coordinate Plot for Hyperparameter Search in Squirrel Dataset