Abstract
Detecting anomalous communities in complex networks is a fundamental task in network science, cybersecurity, and social network analysis. In this paper, we introduce a novel framework for anomalous community detection that accounts for the intentional manipulation of community structures through adversarial deception strategies. Intuitively, our notion of anomalous community captures communities whose structure can be deliberately perturbed to evade community detection algorithms while preserving their internal cohesion. The rationale underlying this novel notion is that well-formed communities are more difficult to hide within the overall network structure and are inherently difficult to conceal, whereas anomalous ones are particularly vulnerable to deception. We formalize the notion of deception anomaly by combining community deception metrics with statistical unexpectedness measures derived from the distribution of deception scores across candidate communities. Building on these definitions, we propose DACAD (Deception-based Adversarial Community Anomaly Detection). We perform a preliminary evaluation of the proposed approach on synthetic benchmark networks, analyzing the effectiveness of the proposed approach for different network topologies and deception budgets and comparing our method with related competing approaches.

This publication has 23 references indexed in Scilit: