Dynamic Network Clustering with Affinity Propagation

A dynamic network clustering algorithm based on Affinity Propagation (AP) is proposed in this paper. The algorithm takes both time consumption and clustering accuracy into consideration. For a node joining, leaving, and moving, procedures are designed respectively to firstlyadjust clusters locallybased on the similarities between the node and the exemplars, and then use AP algorithm globally if the whole redundancy reaches a predefined threshold. Experiments on artificial datasets and KDD99 show that the proposed algorithm reduces time consumption greatly with only less than 5% loss of clustering accuracy.