Targeted attacks based on networks component structure
Issa Moussa Diop, Chantal Cherifi, Cherif Diallo, Hocine Cherifi
Informations
- Revue / Journal
- International Workshop on Complex Networks. Complex Networks XIV. CompleNet 2023. Springer Proceedings in Complexity. Springer, Cham
- Éditeur
- Springer Nature
- Pages
- 62-73
- ISBN
- 978-3-031-28275-1
Résumé
Robustness analysis for targeted attacks is essential, especially for critical infrastructures. Typically, targeted attacks rank the nodes according to a centrality measure and remove top nodes according to a budget. The goal is to exploit the network features efficiently to dismantle them with a minimal budget. Few works are linked to the network mesoscopic properties in the literature, although it is well-admitted that communities or core-periphery are ubiquitous in real-world networks. We propose a network dismantling method based on a mesoscopic representation called the component structure. It performs classical centrality attacks on the network’s global components rather than on the original network. Global components of a network are isolated networks formed by the interactions between its dense parts that one can extract from a community or multiple core-periphery structures. We investigate the proposed strategy using three real-world networks and popular centrality measures (Degree, Betweenness, and PageRank). Results show that the proposed approach is more effective for Degree and PageRank. In contrast, the Betweenness attack on the original network slightly outperforms the attack on the global components but at the price of higher complexity.