Influential Node Ranking in Complex Networks Using A Randomized DynamicsSensitive Approach

Identifying the most influential nodes in information networks has been the focus of many research studies. This problem has crucial applications in various contexts including biology, medicine, and social science, such as controlling the propagation of viruses or rumours in real world networks. While existing methods mostly employ local neighbourhood features which heavily rely on the network structure and disregard the underlying diffusion dynamics, in this work we present a randomized sampling algorithm that not only takes into account the local and global structural features of the network but also considers the underlying diffusion dynamics and its parameters. The main idea is to compute the influentiality of a node through reachability from that node in a set of random graphs. We use a hyper-graph to capture the reachability from nodes in the original network, and theoretically argue that the hypergraph can be used to approximate the theoretical influentiality of nodes in the original graph with a factor of 1 - epsilon. The performance of the proposed model is also evaluated empirically by measuring the correlation between the ranking generated by the proposed method and the ground truth ranking. Our results show that the proposed method substantially outperforms state of the art methods and achieves the highest correlation with the ground-truth ranking, while the generated ranking has a high level of uniqueness and uniformity. Theoretical and practical analysis of the running time of the algorithm also confirms that the proposed method maintains a competitive running time in comparison to the state of the art methods.