Community mining with new node similarity by incorporating both global and local topological knowledge in a constrained random walk

Detection of community is a crucial step to understand the structure and dynamics of complex networks. Most of conventional community detection methods focus on optimizing a certain objective function or on clustering nodes based on their similarities, which leads to a phenomenon that they have preference for specific types of networks but are not general. Using constrained random walk, we exploit global and local topology structures of network to propose a modified transition matrix and further to define a new similarity metric (named ISIM) between two nodes. In contrast to the existing similarities, ISIM does not work directly on the observed data, but in a convergent stable space. This feature makes ISIM robust to the observed noisy data in real-world networks. ISIM not only measures node’s distance, but also captures node’s topology structure in network. Experiments on synthetic and real-world networks demonstrate that ISIM can be successfully applied to community detection in broader types of networks and outperforms other community detection methods.