A hybrid approach to network disintegration: Integrating graph convolutional network and genetic algorithm

The study of network disintegration, which involves objectives such as interrupting rumor-spreading networks and dismantling terrorist networks, has received increasing attention over the past decades. Recent studies have begun employing deep learning architectures to address this problem, but the current training datasets are restricted to small-scale networks due to the exhaustive enumeration required for all possible node combinations, which directly impacts the accuracy and applicability of these learning models to large-scale networks. Here, we propose a new approach that leverages graph convolutional networks for network disintegration by incorporating a genetic algorithm to construct the training dataset; the genetic algorithm can significantly expand the training set to include networks with hundreds or thousands of nodes, thus enhancing the scalability and robustness of the model. Extensive experiments indicate that the neural model trained by the network disintegration strategies from the genetic algorithm significantly outperforms existing methods. Notably, we find that it often requires fewer critical nodes when compared with other methods. The presented framework introduces a new direction for using genetic algorithms in conjunction with deep learning to construct training datasets, which facilitates the disintegration of large-scale networks and improves the accuracy of network disintegration models.