Partitioning of hypergraph modeled complex networks based on information entropy

Abstract

Complex networks with nonuniform degree distribution characteristics are called scale-free networks, which can be divided into several natural imbalanced communities. Hypergraph is good at modeling complex networks, and balanced partitioning. But traditional hypergraph partitioning tools with balance constraints could not achieve good partitioning results for nature imbalanced datasets. In order to partition a complex network into “natural” structure, and reduce the interpart communication cost simultaneously, we make three contributions in this paper. First, we use an information entropy expression considering degree distribution to describe the complex networks. Second, we put forward a partitioning tool named EQHyperpart, which uses complex network information Entropy based modularity Q to direct the partitioning process. Finally, evaluation tests are performed on modern scale-free networks and some classical real world datasets. Experimental results show that EQHyperpart can achieve a tradeoff between modularity retaining and cut size minimizing of hypergraph modeled complex networks.