Feature-Community-guided DICE (FCom-DICE), a perturbation strategy built on DICE that rewires a set of structurally influential edges and adjusts node features to reduce the distinctiveness exploited by GNN message passing is introduced.
Abstract
Graph neural networks (GNNs) enable powerful unsupervised learning of communities. However, such inference may inadvertently expose sensitive group structures, critical clustered patterns, or collective behaviors, raising concerns about sensitive group-level privacy. In social and critical infrastructure networks, unauthorized community inference can reveal coordinated asset groups, operational hierarchies, and system dependencies that may be exploited for reconnaissance or profiling. We study a defensive setting in which a network (or defender) operator seeks to conceal a community of interest while making only small, utility-preserving modifications to the network. Our analysis shows that community concealment depends on two measurable factors: the connectivity at the community boundary and the feature similarity between the protected community and its neighbors. Guided by these observations, we introduce Feature-Community-guided DICE (FCom-DICE), a perturbation strategy built on DICE (Disconnect Internally Connect Externally) that rewires a set of structurally influential edges and adjusts node features to reduce the distinctiveness exploited by GNN message passing. Across synthetic benchmarks and real network graphs such as Facebook, Wikipedia, and Bitcoin Transactions, FCom-DICE consistently outperforms structure-only DICE under the same perturbation budgets. The largest improvements are observed for communities that are weakly connected to the rest of the network and well separated in feature space. These gains are achieved while preserving key structural and feature characteristics of the original network. These results demonstrate the effectiveness of feature-aware perturbations for reducing the recoverability of targeted communities under GNN-based community inference.
Community detection reveals meso-scale structure in graphs but can inadvertently expose sensitive affiliations and operational patterns. We study community structure deception (CSD) as a budgeted edge-perturbation problem on directed networks, aiming to reduce the recoverability of communities under heterogeneous detectors. We present a unified, multilevel framework that operates at three granularities within the same optimization model: structure-wide (CSD), single-community [single-community hiding (SCD)], and individual [individual node hiding (IND)]. Our objective is a direction-aware deception functional built on structural entropy and its directed residual entropy. For CSD and SCD, this objective yields closed-form edge-level differentials and principled edit policies with respect to a selected reference partition. For IND, where the target user has only local ego-network visibility and no partition access, we introduce a label-free ego-centric surrogate based on local degree and neighborhood-overlap statistics. We explicitly distinguish the exact partition-based results from this approximation-based IND heuristic and validate the latter empirically against alternative local scoring rules. Empirically, across multiple detectors (spectral, modularity, stochastic block model (SBM), label-propagation) and datasets, the method consistently lowers detectability by significant margins. Together, these results establish directed residual entropy as a theoretically grounded objective for multilevel community deception with respect to a reference partition and show empirically that the resulting perturbations transfer across heterogeneous detectors, with applications to privacy-preserving data release, robustness benchmarking, and individual anonymity.
Saif Aldeen Madi, Giuseppe Pirrò· IEEE Transactions on Computa...· 0 citations
: Social network platforms have become primary channels for information dissemination, yet they are increasingly exploited by anomalous users such as bots, fake accounts, and coordinated disinformation spreaders. These malicious actors manipulate public opinion, spread misinformation and undermine platform integrity, posing severe threats to the security of the online ecosystem. Accurate detection of such users is challenging because they often organize into sophisticated high-order connection patterns that extend beyond local neighborhoods. Existing methods address this by either injecting predefined motifs as handcrafted features, which lack flexibility to discover unknown patterns, or employing higher-order Graph neural networks (GNNs) at prohibitive costs. Crucially, neither method treats structural information as learnable knowledge that can be automatically acquired from data and explicitly represented. To bridge this gap, we propose SparseGNN, a structural-knowledge-enhanced framework for anomalous user detection. It regards atomic subgraph patterns as fundamental, learnable units of structural knowledge. This framework is concatenated with original node features and fed into any standard GNN, without modifying the backbone architecture. Experiments on real-world datasets demonstrate that SparseGNN improves the accuracy and F1-score of standard GNNs for anomalous users detection without requiring predefined patterns, while maintaining linear complexity. Because the learned atomic patterns capture global high-order topology, the resulting structural knowledge representation is inherently less sensitive to localized edge perturbations, incidentally conferring improved stability under adversarial structural attacks.
Zehan Li, Yingyi Li, Zhiwei Tang et al.· Computers, Materials & C...· 0 citations
Many real-world networks, such as social and biological networks, exhibit community structures. Community detection algorithms extract valuable insights from these networks by identifying densely connected groups, enabling applications such as recommendation, behavior understanding, and system optimization. However, growing concerns about data privacy and security have led to techniques that protect user information from being over-inferred within communities. This has given rise to community deception (CD), which introduces small, targeted perturbations to a network to obscure sensitive communities from detection algorithms. Most existing community deception approaches focus on modifying network topology, often neglecting the rich feature information embedded within communities. In this paper, we propose FSC-CD (Feature-Structure Coupled Community Deception), which couples feature-derived representations with structural cues to improve community concealment. FSC-CD is effective for both single-community deception and randomized multi-community hiding. A key innovation is a budget allocation strategy that optimizes the distribution of perturbations to maximize deception efficiency. Moreover, by exploiting feature similarity, FSC-CD designs an edge perturbation mechanism that improves stability under small perturbation budgets. Extensive experiments on three real-world network datasets across multiple community detectors show that FSC-CD is more stable and consistently outperforms baseline methods in hiding both single and multiple communities, reducing the detection accuracy by up to 17.6 % compared to state-of-the-art approaches.
Yue Yang, Zhuoyan Yang, Yutong Liu et al.· International Conferences on...· 0 citations
Graph embeddings offer a principled way to discover coordinated attack patterns in security telemetry, but their hyperparameters are typically inherited from social-network benchmarks without validation. We conduct a systematic sensitivity analysis of Node2Vec on a heterogeneous security event graph derived from production AWS cloud telemetry (237 entities, 500+ relationships), varying six hyperparameters across 81 experimental configurations. Our central finding: the number of random walks dominates all other parameters (composite importance 1.00) through a non-monotonic relationship: clustering quality peaks sharply at r=100 (silhouette ≈0.56) then degrades to 0.27 at r=300. We attribute this to over-smoothing: in sparse, heterogeneous graphs, excessive walks cause node representations to collapse toward a mean, destroying the discriminative structure that separates attack phases. By contrast, the biased-walk parameters p and q, which receive the most practitioner attention, rank last in importance. Four recovered clusters map cleanly to MITRE ATT&CK kill chain phases. These results challenge the "more sampling is better" assumption and provide actionable, empirically-grounded configuration guidance for practitioners deploying graph-based threat detection on investigation-scale security graphs.
Moazzam Khan, A. W. Malik, Mohammad-Ali Omer· 2026 IEEE Cloud Summit· 0 citations
Community detection aims to uncover mesoscopic organization in graphs by grouping nodes into densely connected communities, and it plays an important role in analyzing citation networks, social systems, and other complex relational data. In practice, this task is challenging because community structures are determined not only by pairwise connectivity, but also by higher-order structural patterns such as motifs. Existing Graph Masked Autoencoders (GMAEs), although effective for self-supervised graph representation learning, usually rely on random node or edge masking and pairwise reconstruction objectives. As a result, they may fail to preserve the higher-order structural regularities that are crucial for identifying coherent communities. To address this issue, we propose a Motif-aware Graph Masked Autoencoder (MGMAE) for community detection. The proposed framework explicitly incorporates triangle motifs into both the masking and reconstruction processes. Specifically, MGMAE first enumerates motif structures as higher-order supervision targets, then constructs perturbed graphs through a joint motif–edge masking strategy, and finally learns node representations by jointly reconstructing masked edges and masked motifs. In this way, the learned embeddings are encouraged to preserve both local connectivity and higher-order cohesion patterns that are closely related to community organization. Extensive experiments on six real-world attributed networks and five synthetic LFR benchmarks demonstrate that MGMAE achieves highly competitive performance against representative baselines in terms of NMI and ARI. Additional ablation and hyperparameter studies further verify the effectiveness of the proposed motif-aware masking and reconstruction design.
Chuanpeng Wang, Dong Li, Ruyi Qiu et al.· Journal of King Saud Univers...· 0 citations
The application of graph data in numerous disciplines raises the need for gathering and analyzing huge volumes of data, some of which is private and sensitive. The non-Euclidean nature of the graph data makes the analysis computationally challenging, leading to the use of Graph Neural Networks (GNNs) in the age of AI. GNNs may inadvertently leak sensitive data they are trained on, which raises serious data security issues, including the model inversion attack. In this study, we analyze GNNs'vulnerabilities by introducing two novel graph inversion (i.e., reconstruction) attacks: graph-label conditioned (GLC) attack and embedding-label conditioned (ELC) attack, utilizing targetmodel predictions and their intermediate representations, respectively. We perform a comprehensive analysis of our introduced privacy attacks and compare them with existing baselines across three benchmark graph datasets (i.e., NCI1, PROTEINS, and AIDS) and four graph distributional/structural metrics (i.e., FGD, EGD, MMD, and GKS). Our work demonstrates that an adversary can use the generator-discriminator technique to reconstruct high-quality graphs in real-world black-box attack scenarios against GNNs. Additionally, we present a variant of our attacks (Ours--) with 50% reduced queries, achieving good or comparable reconstruction attack performance. In addition, we show that GNNs are highly vulnerable to privacy attacks, varying Laplacian noise-scales.
A. Keji, S. Dibbo· arXiv.org· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduAug 18, 2026
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
