Mining important nodes in complex networks using nonlinear PCA

摘要

An important problem in the analysis of complex networks is to mine the top k important nodes. The existing literature offers several metrics, also called centrality measures which estimates importance using the structural properties of node, namely, degree, closeness, betweenness, eigenvector centrality, Pagerank etc. Though there exists plenty of centrality measures, none of them emphasizes the non-linearity of data. In the current study we propose a non-linear principle component analysis based approach to identify the top k important nodes. The proposed method is evaluated based on the saturation time and the fraction of infected nodes during a susceptible-infected propagation. The experiment on synthetic as well as real life data sets show that the developed method is competitive with the state-of-the-art.