Date of Graduation


Document Type


Degree Name

Bachelor of Science in Computer Engineering

Degree Level



Computer Science and Computer Engineering


Zhang, Lu

Committee Member/Reader

Gauch, Susan

Committee Member/Second Reader

Patitz, Matthew


Message-passing graph neural networks (MPGNNs) are known to have limitations in their representational power. Recent work proposes subgraph graph neural network (subgraph GNN) models to address these limitations by upgrading the local node representations of MPGNNs to respective subgraph representations. However, existing subgraph GNN models have limited interpretability in capturing inherent local structural dynamics across diverse graph structures. In this work, we present a novel subgraph GNNs framework, termed Community-Induced Graph Neural Network (CiGNN). The key idea of CiGNN is to endow an intuitive interpretability basis for subgraph GNNs by capturing the dynamics of inherent structural community topology in subgraph representation realization relative to individual nodes. Based on this idea, we propose four variants of CiGNN in establishing graph representations from community-induced local subgraphs. Specifically, we approach subgraph formulation from a statistical physics perspective combined with network analysis. We leverage a theoretical basis from the Reichardt and Bornholdt's Potts (RBP) model and Louvain-based community optimization methods to consider the combination of two null model initializations for the RBP models and two community optimization methods. Our modular framework supports multiple mainstream GNN base models, making CiGNN a general module easily adaptable to existing GNN or subgraph GNN models with intuitive interpretability. Experimental results on six benchmark graph classification datasets show that CiGNN achieves competitive performance against subgraph GNN baselines and recent state-of-the-art GNNs.


Graph Neural Network, Graph Classification, Representational Learning