.png){kind=icon}
.png){kind=icon}
Return
Complex network dismantling with node diffusion and node similarity
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.130433.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Network dismantling (ND) achieves the disintegration of an entire network by selecting a set of key nodes to decompose it into disconnected subgraphs. Identifying the minimal target attack set (TAS) of critical nodes for ND constitutes an NP-hard problem. Recent studies have demonstrated that ND-oriented neural networks outperform traditional approaches. However, existing methods still face two major limitations: (i) inadequate capture of node information and (ii) excessive computational overhead. To address these challenges, this paper proposes an enhanced methodology for efficient critical node identification and rapid network dismantling. Our research reveals that similarity assessment can serve as a topological evaluation of nodes, while propagation capability assessment acts as a key metric for measuring information transmission efficiency. Based on this observation, the paper infers that nodes achieving high scores in both the topological and propagation dimensions possess superior network dismantling capability. Consequently, we propose a scoring framework that evaluates both the node propagation capability and the node similarity. On one hand, a multi-head attention mechanism is employed to rapidly assess node propagation capability. On the other hand, an unsupervised community partition method, combined with multi-view contrastive learning, is used to measure node similarity. In addition, the loss function is optimized by prioritizing the loss of the dismantling task to refine the scoring process. This paper conducted extensive experiments on both real-world and synthetic networks. The results demonstrate that our method achieves faster network dismantling with fewer nodes compared to other baselines.
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
Organization
No organization information available