Towards Robust Graph Structural Learning Beyond Homophily via Preserving Neighbor Similarity

Unknown Date

AI Summary

This work addresses the vulnerability of graph learning models to adversarial attacks and insufficient structural robustness on both homophilous and heterophilous graphs. We propose a novel robust graph structure learning paradigm that preserves neighbor similarity. Theoretically, we prove that negative classification loss updates are inversely correlated with neighbor feature similarity under power-iteration aggregation. Leveraging this insight, we design a dual-kNN-guided adaptive propagation mechanism enabling topology-robust mining across graph types. Our method comprises: (i) dual-kNN graph construction, (ii) similarity-constrained propagation, (iii) adaptive graph convolution with dynamic switching between smoothing and discriminative modes, and (iv) power-iteration-based neighbor feature aggregation. Extensive experiments on diverse homophilous and heterophilous benchmarks demonstrate significant improvements in adversarial robustness, more reliable topology reconstruction, an average 5.2% gain in classification accuracy, and superior generalization over existing state-of-the-art methods.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Game Theory and Economic Paradigms: Adversarial Learning

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: Robustness and generalizability of Web mining methods

Abstract

Despite the tremendous success of graph-based learning systems in handling structural data, it has been widely investigated that they are fragile to adversarial attacks on homophilic graph data, where adversaries maliciously modify the semantic and topology information of the raw graph data to degrade the predictive performances. Motivated by this, a series of robust models are crafted to enhance the adversarial robustness of graph-based learning systems on homophilic graphs. However, the security of graph-based learning systems on heterophilic graphs remains a mystery to us. To bridge this gap, in this paper, we start to explore the vulnerability of graph-based learning systems regardless of the homophily degree, and theoretically prove that the update of the negative classification loss is negatively correlated with the pairwise similarities based on the powered aggregated neighbor features. The theoretical finding inspires us to craft a novel robust graph structural learning strategy that serves as a useful graph mining module in a robust model that incorporates a dual-kNN graph constructions pipeline to supervise the neighbor-similarity-preserved propagation, where the graph convolutional layer adaptively smooths or discriminates the features of node pairs according to their affluent local structures. In this way, the proposed methods can mine the ``better" topology of the raw graph data under diverse graph homophily and achieve more reliable data management on homophilic and heterophilic graphs.

Problem

• Enhancing adversarial robustness on heterophilic graphs
• Addressing vulnerability beyond homophilic graph data
• Preserving neighbor similarity across diverse homophily degrees

Innovation

• Dual-kNN graph construction for neighbor similarity preservation
• Adaptive feature smoothing based on local structures
• Robust graph mining across homophilic and heterophilic graphs