Graph variational autoencoder with affinity propagation for community-aware anomaly detection in attributed networks

Anomaly detection in attributed networks (ADAN) aims to identify abnormal nodes that exhibit unexpected link structures and attributes compared to the others. The existing works primarily utilize the node representations learned from the low-level attributes and link structures of nodes to detect ab...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Applied soft computing Jg. 186; S. 114223
Hauptverfasser: Cao, Zhijie, Yang, Chengkun, Fan, Xiaoqing, Li, Lingjie, Lin, Qiuzhen, Li, Jianqiang, Ma, Lijia
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 01.01.2026
Schlagworte:
Affinity propagation
Anomaly detection
Attributed network
Community detection
Graph variational autoencoder
Attributed network
Community detection
Graph variational autoencoder
Anomaly detection
Affinity propagation
ISSN:1568-4946
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Anomaly detection in attributed networks (ADAN) aims to identify abnormal nodes that exhibit unexpected link structures and attributes compared to the others. The existing works primarily utilize the node representations learned from the low-level attributes and link structures of nodes to detect abnormal nodes, which overlook the impact of high-level community structures on ADAN. To address this issue, this paper proposes a novel framework called Graph Variational Autoencoder with Affinity Propagation (GVE-AP) for community-aware ADAN. GVE-AP first employs a graph convolutional variational autoencoder to learn node embeddings from attributed networks. Then, it integrates an affinity propagation algorithm for community division, which jointly considers both node attributes and link structures. Subsequently, it introduces a novel community-aware anomaly score to detect abnormal nodes by measuring dissimilarity with their communities based on robust features extracted via principal component analysis. Experimental results on eight real-world datasets demonstrate that GVE-AP outperforms the state-of-the-art methods for anomaly detection in attributed networks in terms of AUC and robustness. •Proposed a method to learn low-dimensional node representations using link and attribute information.•Developed a mechanism to detect communities based on structural and attribute similarities.•Designed an anomaly score calculation integrating link structure, node attributes, and communities.•The proposed model outperforms other methods in anomaly detection performance.
ISSN:1568-4946
DOI:10.1016/j.asoc.2025.114223