Unveiling Anomalous Edges and Nominal Connectivity of Attributed Networks

TL;DR

This paper introduces two efficient methods for detecting anomalous edges in attributed networks, improving robustness and accuracy in identifying outliers and adversarial connections across various applications.

Contribution

It proposes two novel, complementary formulations for anomaly detection in attributed graphs, emphasizing efficiency and robustness in uncovering diverse anomalous edges.

Findings

Effective detection of anomalous edges linking different communities

Robust recovery of unperturbed graph enhances anomaly identification

Validated on real and synthetic datasets

Abstract

Uncovering anomalies in attributed networks has recently gained popularity due to its importance in unveiling outliers and flagging adversarial behavior in a gamut of data and network science applications including {the Internet of Things (IoT)}, finance, security, to list a few. The present work deals with uncovering anomalous edges in attributed graphs using two distinct formulations with complementary strengths, which can be easily distributed, and hence efficient. The first relies on decomposing the graph data matrix into low rank plus sparse components to markedly improve performance. The second broadens the scope of the first by performing robust recovery of the unperturbed graph, which enhances the anomaly identification performance. The novel methods not only capture anomalous edges linking nodes of different communities, but also spurious connections between any two nodes with different features. Experiments conducted on real and synthetic data corroborate the effectiveness of both methods in the anomaly identification task.