Research On Community Detection Algorithm Based On Complex Network

Network is an effective way to present complex systems. Complex network describes aseries of natural and social systems, such as information systems, biological systems, socialsystems, transportation systems, aviation network, power systems, and some cooperativenetwork systems. A lot of networks in nature have unique structural features, the relationshipand the interaction between tuples in the system, can be abstracted as a small network with acertain organizational relationships, which serves function of the system and concretelyreflects organizational relationships of the network. These small networks can be defined asthe community structure of complex network.The research of community detection of complex network can reveal the structuralfeatures of complex network and some causes of networking phenomenon. To divideindividuals of complex network into different community structure to study, can studytargetedly certain features of complex network, and can also reduce the workload of study ofsome features of complex network. Therefore, the study of community detection methods forcomplex network have enormous social, economic and scientific value.Focusing on complex network, this paper firstly proposed community-similarity-basedhierarchical clustering community detection algorithm CSHC, which was inspired byLouvain community detection algorithm. The initial phase of the algorithm regards each nodeas a community, and then proposes a merger coefficient based on the similarity ofcommunities and maximum gain of modularity, which decide whether to merge thecommunity, the stopping criterion for iterative algorithm is that the number of communities isequal to the desired number k. CSHC algorithm can get very good community division resulton both Karate Club Network dataset and American College Football dataset, its purity andmodularity have been improved to a certain extent compared with the past communitydetection algorithms. CSHC algorithm can get community division result with small times ofmerge, and its efficiency has been greatly increased compared with the conventionalcommunity detection algorithms. Secondly, this paper proposed an important-node-basedcommunity detection algorithm INC, first of all based on modularity maximization theory,calculate the maximum k eigenvectors matrix S of the modularity matrix B of the network.Then, the method of cluster centrality was proposed, which calculates the important nodes ofthe k communities and regards these nodes as k cluster centralities. It uses Euclidean distanceto calculate the distance between each node and k cluster centralities, and assign the node tothe community where the nearest cluster centrality is in. Finally, apply k-means iterativecalculation method to the network and eventually get the k community divisions. INCalgorithm can get very good community division result on both Karate Club Network datasetand American College Football dataset and can effectively identify potential community, itspurity and modularity have been improved to a certain extent compared with the pastcommunity detection algorithms. INC algorithm can get community division result withsmall times of merge, and its efficiency has been greatly increased compared with the conventional community detection algorithms.The structures of various community within complex network is the concrete undertakerof structural features of complex network and the concrete embodiment of attribute featuresof complex network. In addition, not all the tuples of the individual play the same role in thefeatures and structures of complex network, some tuples may play a more important role,these tuples are called key nodes of the network. Therefore, this paper proposed importantnode assessment method based on community, it both takes full account of the closenessbetween node and all other nodes, but also takes full account of the contribution of node inthe community, this paper proposed ICC algorithm to access the importance of nodes innetwork. ICC algorithm was verified by experiments in Karate Club Network dataset andAmerican College Football dataset respectively, and was compared with the classicalcentrality calculation methods. The experimental results show that ICC algorithm can wellhighlight the practical significance of network that importance nodes play, provides a newperspective of the evaluation of the important nodes in network, and plays a good judge onthe practical significance of network.The modern scientific network brings has brought significant development of theunderstanding of complex network. The study of complex network community detection andcommunity detection algorithm plays an important role in the discussion of key issues, themeaning of clusters and the description of real network. Because each node of complexnetwork has different importance, each node has different contribution to the attributes ofcomplex network. Therefore, mining key nodes in a complex network has great practicalvalue.