Exploring Heterophily in Graph-level Tasks

2025-09-23

AI Summary

This work investigates the impact mechanism of heterophily (disassortativity) in graph-level tasks, addressing a critical gap in both theoretical understanding and methodological development for graph-level learning. We propose a graph-level label pattern taxonomy, grounded in motif-based local structural signatures, and theoretically establish—via spectral analysis—for the first time that motif detection relies on a dynamic mixture of multiple spectral components, contradicting conventional global-frequency-dominant mechanisms. Building on this insight, we design a frequency-adaptive GNN architecture and conduct rigorous theoretical analysis based on energy gradient flow. Experiments on synthetic benchmarks and real-world molecular property prediction tasks demonstrate that our model significantly outperforms frequency-dominant baselines under controlled heterophilous settings, empirically validating the essential role of spectral adaptivity in modeling graph-level heterophily.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

While heterophily has been widely studied in node-level tasks, its impact on graph-level tasks remains unclear. We present the first analysis of heterophily in graph-level learning, combining theoretical insights with empirical validation. We first introduce a taxonomy of graph-level labeling schemes, and focus on motif-based tasks within local structure labeling, which is a popular labeling scheme. Using energy-based gradient flow analysis, we reveal a key insight: unlike frequency-dominated regimes in node-level tasks, motif detection requires mixed-frequency dynamics to remain flexible across multiple spectral components. Our theory shows that motif objectives are inherently misaligned with global frequency dominance, demanding distinct architectural considerations. Experiments on synthetic datasets with controlled heterophily and real-world molecular property prediction support our findings, showing that frequency-adaptive model outperform frequency-dominated models. This work establishes a new theoretical understanding of heterophily in graph-level learning and offers guidance for designing effective GNN architectures.

Problem

• Analyzing heterophily's impact on graph-level learning tasks
• Investigating motif detection requirements for mixed-frequency dynamics
• Designing frequency-adaptive GNN architectures for graph-level tasks

Innovation

• Analyzed heterophily in graph-level tasks theoretically and empirically
• Used energy-based gradient flow for motif detection analysis
• Proposed frequency-adaptive models outperform frequency-dominated architectures