Community detection via measuring the strength between nodes for dynamic networks

The detection of community structure for dynamic social networks is significant for understanding evolution features of collective behaviors. In this paper, we present community detection method based on nonnegative matrix factorization for dynamic networks considering the strength between nodes. The basic idea of this algorithm is that node pairs with stronger connection strength have more possibility to be grouped into the same community. Firstly, we build weighted networks by calculating the embeddedness Et and dispersion Dt between each pair of nodes to measure the strength of the relationships at each timestamp t. Then we construct a node strength matrix in which each element represents the connection strength of a pair of nodes. Combining the structural information at previous timestamp, the nonnegative matrix factorization method is used to detect the community structure for the dynamic networks. Finally, the experiments for two synthetic networks show that when considering the previous information, the accuracy of our algorithm improve 0.3425, 0.5191 for the first synthetic networks. For the second synthetic networks, the accuracy of our algorithm is also improved. Furthermore, we compare the other two algorithms, the results show that our algorithms perform better than other algorithms on the both synthetic networks. Our work may be helpful for providing a new perspective that we detect community structures for dynamic networks.