Discovering Invariant Neighborhood Patterns for Heterophilic Graphs

2024-03-15

arXiv.org

AI Summary

In heterogeneous graphs, neighborhood pattern distribution shifts—where node connections significantly violate the homophily assumption—severely degrade the generalization of existing Graph Neural Networks (GNNs). Method: This paper proposes a synergistic framework of Adaptive Neighborhood Propagation and Invariant Graph Representation Learning. It introduces a novel neighborhood-pattern invariance learning paradigm: the Adaptive Neighborhood Propagation (ANP) module dynamically models heterogeneous neighborhood structures, while the Invariant Neighborhood-aware Graph Learning (INHGL) module explicitly enforces node representation invariance across distributions. The method operates without relying on the homophily assumption, ensuring both distributional robustness and scalability. Contribution/Results: Evaluated on multiple large-scale real-world heterogeneous graph benchmarks, the approach achieves state-of-the-art performance, substantially improving model generalization and robustness under distribution shifts. It establishes a new paradigm for learning on non-homophilous graphs.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Game Theory and Economic Paradigms: Adversarial Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

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
• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks

Abstract

This paper studies the problem of distribution shifts on non-homophilous graphs Mosting existing graph neural network methods rely on the homophilous assumption that nodes from the same class are more likely to be linked. However, such assumptions of homophily do not always hold in real-world graphs, which leads to more complex distribution shifts unaccounted for in previous methods. The distribution shifts of neighborhood patterns are much more diverse on non-homophilous graphs. We propose a novel Invariant Neighborhood Pattern Learning (INPL) to alleviate the distribution shifts problem on non-homophilous graphs. Specifically, we propose the Adaptive Neighborhood Propagation (ANP) module to capture the adaptive neighborhood information, which could alleviate the neighborhood pattern distribution shifts problem on non-homophilous graphs. We propose Invariant Non-Homophilous Graph Learning (INHGL) module to constrain the ANP and learn invariant graph representation on non-homophilous graphs. Extensive experimental results on real-world non-homophilous graphs show that INPL could achieve state-of-the-art performance for learning on large non-homophilous graphs.

Problem

• Addressing distribution shifts in non-homophilous graphs
• Learning invariant neighborhood patterns for heterophilic graphs
• Improving graph neural networks for complex real-world graphs

Innovation

• Invariant Neighborhood Pattern Learning for graphs
• Adaptive Neighborhood Propagation module
• Invariant Non-Homophilous Graph Learning module