Community Detection in Graph Streams by Pruning Zombie Nodes

Abstract

Detecting communities in graph streams has attracted a large amount of attention recently. Although many algorithms have been developed from different perspectives, there is still a limitation to the existing methods, that is, most of them neglect the “zombie” nodes (or unimportant nodes) in the graph stream which may badly affect the community detection result. In this paper, we aim to deal with the zombie nodes in networks so as to enhance the robustness of the detected communities. The key here is to design a pruning strategy to remove unimportant nodes and preserve the important nodes. We propose to recognize the zombie nodes by a degree centrality calculated from the exponential time-decaying edge weights, which can be efficiently updated in the graph stream case. Based on only important and active nodes, community kernels can be constructed, from which robust community structures can be obtained. One advantage of the proposed pruning strategy is that it is able to eliminate the effect of the aforementioned “zombie” nodes, leading to robust communities. By designing an efficient way to update the degree centrality, the important and active nodes can be easily obtained at each timestamp, leading to the reduction of computational complexity. Experiments have been conducted to show the effectiveness of the proposed method.