An Evolving Network Model Based On Interest Vectors And Dynamic Benchmark Graphs

Researching on complex networks is the entry point of complex systems,and also helps humans understanding the robustness,evolving pattern and information propagation of complex networks.On the one hand,modeling is the main way to describe the complex network.On the other hand,splitting the network into communities could simplify its complexity,and revealing its dynamical principle of the formation of communities.This paper proposes three outcomes:First,an evolving network model based on interest vectors is proposed,in which the formation of communities is driven by each node's various preference to fields of interest.The very intrinsic properties of nodes are denoted as interest vectors,of which the mechanism is far different from other existing models with community structure.Second is a fast dynamic benchmark network algorithm,which could generate a series of snapshots of a dynamic network.These snapshots record edges and vectors of the network,as well as community information and memberships of nodes.Benchmark networks usually generate simulated data mimicking real-life networks.Applying these data to test community detection algorithms could evaluate these algorithms' speed and accuracy.In contrast with existing benchmark networks,the new dynamic benchmark could produce dynamic information about a network,instead of static.The dynamic benchmark is tested by community algorithms and compared with a popular static algorithm.Third,an extensible framework for generating evolving graphs is proposed,for researchers to implement their models easily leveraging on the framework.Due to the commonality of the preferential attachment,a data structure named Preferential Attachment Pool is introduced,supporting weight computing,with low time and space complexity.The efficient data structure makes it possible to turn evolving models into benchmark graphs.In the theoretical analysis and numerical simulation,the model is shown that the degree distribution of the model represents a transition between power-law and exponential,depending on parameters given,which is close to real-life networks.For benchmark networks,multiple algorithms are performed on the graph,then evaluate the results.Finally,it is proved that the benchmark graphs well distinguish different algorithms.