Study Of Community Detection Algorithms Based On Node Similarity

With the ongoing development of complex networks,the detection requirements of network structure become more and more stringent.Therefore,community detection algorithms have been proposed and improved continuously.In the real network,the distributions of nodes and edges are nonuniform,and there are always some places which are dense or sparse.In the community,dense local networks appear as the community structure,whereas sparse edges are the boundaries of different communities.Link prediction theory is a kind of study method for network to predict the possibility of a connection among unconnected nodes in the network by using the internal structure of the network,the node and even the edge and other information.The similarity between nodes is an important measure of the relationship between two nodes,because the greater the similarity between the two nodes,the more likely that the two nodes are connected.On the basis of the link prediction theory of node similarity,a kind of community detection algorithm for non-overlapping networks has been proposed in this paper.The algorithm acquires community numbers and core node sets automatically,as a result of the completion for the network lost information through link prediction and the division of network redundancy information through removing the link.Then,label propagation processing have been finished for isolated nodes to achieve network clustering eventually.Two kinds of community detection algorithms based on node similarity have been proposed firstly and elaborated at length in the paper.They are ADL-CN algorithm based on common neighbor index(CN index)and ADL-RA algorithm based on the index for the sum of reciprocal about the degree of common neighbors(RA index).Then,in terms of the shortcomings of the above algorithm,ADL-CNRA algorithm combined with them(CN index and RA index)has been proposed.Finally,the three algorithms mentioned in this paper are compared with two classical algorithms in the data set,which verified that the algorithm proposed in this paper has better effect and the results are superior to the comparable algorithms.