Research On Community Evolution Tracking Algorithm In Complex Networks

With the development of information age,a large amount of data is generated all the time,and we are confronted with the explosive growth of data.How to effectively use and mine these implied data is the concern of current scholars.Complex networks,as a basic discipline,appearing in the field of data mining,is often used to solve the related problems of complex systems.Community structure is an important characteristic of real complex network.It can reflect some structural characteristics of complex networks,helping us understand complex networks deeper,and mine implicit information of the complex network.The theory of community structure in complex network is also applied to solve practical problems in other fields,such as anomaly detection in social networks,recommendation system based on scoical interaction and so on.So the mining and tracking of the community structures for complex networks have also become a hot topic in the field of data mining.It is found that because the structure of complex networks is changed over time,community structure is not constant but dynamic.In addition,because of the complexity of a relationship,a node often bears multiple roles in a complex network,so the structure of the community is not independent of each other,but maybe overlapping each other.Considering of the characteristics of dynamic change over time and the phenomenon of overlapping of community structure,this paper proposes two evolution frameworks for dynamic networks,namely,the evolutionary framework of non overlapping communities in dynamic networks and the evolutionary framework of overlapping communities in dynamic networks.The innovative results of this article are as follows:1)We propose an evolutionary framework model for non overlapppering community detection in dynamic network.The framework includes two algorithms: One is a static non overlapping community partition algorithm CGEC based on core node subgraph expansion,and the other is a dynamic non overlapping community evolution algorithm DCIC based on incremental clustering.At the initial time,we use the CGEC algorithm to get the initial community structure,and use the DCIC algorithm to incrementally cluster the local network area at the subsequent time,so that we can quickly track the non overlapping community evolution trajectory of large-scale dynamic network.By comparing experiments on artificial and real data sets,it is proved that the framework has high accuracy and efficiency under largescale dynamic network.2)We propose an evolutionary framework model for overlapppering community detection in dynamic networks.The framework includes two algorithms: One is the static overlapping community partition algorithm CLK based on the extension of the core link subgraph,and the other is the dynamic overlapping community evolution algorithm based on incremental clustering DCLK.At the initial time,we use the CLK algorithm to get the initial community structure.Then we use the DCLK algorithm to incrementally cluster the local network area,so that we can capture the evolution path of the overlapping communities of the dynamic network.By comparing experiments on artificial and real data sets,it is proved that the framework has a high accuracy.