Community Detection Based On Embedded Vector And Label Propagation

Complex networks exist in every aspect of real life and grow in scale as society advances.Within a system which varies from the tiny interacted system of huge number cells in an organism to the huge social circle composed of thousands of individuals or even countries,the relationship between the individual and the whole can be expressed in the form of complex networks.Therefore,complex networks are becoming more and more attractive.In complex networks,community structure is one of the most important features that can solve many practical problems.Community structure can not only express the local feature of nodes’ behavior in the network,but also reflect the mutual relationship among nodes.Researchers began to study the community structure in complex networks as early as the last century.And there have been a lot of good works.The study of community structure can help us to understand the structure and function of complex networks,and it plays a crucial role in the analysis and prediction of the interaction among network nodes.Recently years,researchers have proposed a large number of community detection algorithms from different perspectives.The label propagation algorithm is one of the most classical algorithms.The idea of label propagation is simple and it’s easy to implement with low complexity and efficient execution,so it gets a lot of attention.However,the label propagation algorithm has an obvious shortcoming,the update of nodes’ label is unstable in every iteration,which leads to great difference in the result of community partition.In the research on complex networks,network representation learning is a very important technology,it represents networks by the distributed method.Each node in the network is represented in the form of a vector.The relationship among the vectors can reflect the important relationship among the nodes.For getting these vectors,the key is the walk strategy.Recently years,the algorithms of walk strategy have been proposed continuously,node2 vec is one of them.Its strategy is very novel,which combines BFS and DFS to explore fully the structural feature of the network and the relationship among nodes.In this paper,we combine the embedded vector and the label propagation algorithm,then calculate the similarity among nodes in the network through vectors trained by node2 vec.In addition,the homophily and structural equivalence amongnodes in the network can be taken into consideration and expressed in the vectors by controllable parameters in the node2 vec method.The proposed algorithm is tested on several open standard data sets and compared with some existing community detection algorithms.The experimental results show that the algorithm is more efficient and performs better on community partition.