GJDNet: Robust Graph Neural Networks via Joint Disentangled Learning Against Adversarial Attacks

Graph Neural Networks (GNNs) are vulnerable to adversarial attacks, which inherently invert connectivity patterns by introducing disassortative edges in assortative graphs and assortative edges in disassortative graphs. This structural inversion creates structure-feature mismatches that disrupt neighborhood aggregation across different graph types. However, we find that existing defenses are limited, as they either treat neighborhoods as monolithic under fixed assortativity assumptions or rely on standard softmax classifiers that fail to account for perturbation-induced representation shifts. To further exploit this observation, we adopt a robustness perspective that jointly disentangles node representations and decision spaces, isolating perturbation effects while enforcing well-separated decision regions. Based on this principle, we propose Graph Joint Disentanglement Network (GJDNet), a unified framework for robust node classification across diverse graph assortativity regimes. GJDNet enhances robustness at both representation and decision levels: it employs feature-driven soft structural disentanglement with skewness-aware neighbor filtering to suppress perturbation-induced structure-feature mismatches, and introduces a Spherical Decision Boundary (SDB) to promote intra-class compactness and inter-class separation in the embedding space, thereby stabilizing decision boundaries under perturbations. Theoretical analysis provides insights into the effectiveness of the proposed disentangled representation and decision mechanisms, while extensive experiments demonstrate that GJDNet consistently achieves strong robustness across graphs with different connectivity regimes.

The maximum capability of a topological feature in link prediction

Link prediction aims to predict links of a network that are not directly visible, with profound applications in biological and social systems. Despite intensive utilization of the topological feature in this task, it is unclear to what extent a particular feature can be leveraged to infer missing links. Here, we show that the maximum capability of a topological feature follows a simple mathematical expression, which is independent of how an index gauges the feature. Hence, a family of indexes associated with one topological feature shares the same performance limit. A feature's capability is lifted in the supervised prediction, which in general gives rise to better results compared with unsupervised prediction. The universality of the pattern uncovered is empirically verified by 550 structurally diverse networks, which can be applied to feature selection and the analysis of network characteristics associated with a topological feature in link prediction.