A Better Strategy of Discovering Link-Pattern Based Communities by Classical Clustering Methods

The definition of a community in social networks varies with applications. To generalize different types of communities, the concept of link-pattern based community was proposed in a previous study to group nodes into communities, where the nodes in a community have similar intra-community and inter-community interaction behaviors. In this paper, by defining centroid of a community, a distance function is provided to measure the similarity between the link pattern of a node and the centroid of a community. The problem of discovering link-pattern based communities is transformed into a data clustering problem on nodes for minimizing a given objective function. By extending the partitioning methods of cluster analysis, two algorithms named G-LPC and KM-LPC are proposed to solve the problem. The experiment results show that KM-LPC outperforms the previous work on the efficiency, the memory utilization, and the clustering result. Besides, G-LPC achieves the best result approaching the optimal solution.