Adapting to Heterophilic Graph Data with Structure-Guided Neighbor Discovery

2025-06-10

AI Summary

Graph Neural Networks (GNNs) suffer performance degradation on heterophilous graphs due to the failure of the local homophily assumption. Method: This paper proposes a structure-guided neighborhood discovery paradigm: it leverages node structural role embeddings to capture global and local structural similarities, thereby constructing a high-homophily auxiliary graph to reconstruct message-passing pathways; theoretically proves that reducing false-positive edges improves the generalization bound; and introduces a multi-view GNN with adaptive graph-weight learning for effective multi-graph fusion. Contribution/Results: This is the first work to jointly model structural roles and perform graph reconstruction for heterophilous learning. It achieves state-of-the-art performance across multiple heterophilous benchmarks, significantly outperforming standard GNNs and existing heterophily-adapted methods.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Bridging structured and unstructured data

Abstract

Graph Neural Networks (GNNs) often struggle with heterophilic data, where connected nodes may have dissimilar labels, as they typically assume homophily and rely on local message passing. To address this, we propose creating alternative graph structures by linking nodes with similar structural attributes (e.g., role-based or global), thereby fostering higher label homophily on these new graphs. We theoretically prove that GNN performance can be improved by utilizing graphs with fewer false positive edges (connections between nodes of different classes) and that considering multiple graph views increases the likelihood of finding such beneficial structures. Building on these insights, we introduce Structure-Guided GNN (SG-GNN), an architecture that processes the original graph alongside the newly created structural graphs, adaptively learning to weigh their contributions. Extensive experiments on various benchmark datasets, particularly those with heterophilic characteristics, demonstrate that our SG-GNN achieves state-of-the-art or highly competitive performance, highlighting the efficacy of exploiting structural information to guide GNNs.

Problem

• GNNs struggle with heterophilic graph data
• Improving GNNs by reducing false positive edges
• Enhancing GNN performance using multiple graph views

Innovation

• Linking nodes with similar structural attributes
• Utilizing multiple graph views for better performance
• Adaptively weighing contributions of different graph structures