Spectrum Based Clique Expansion

In our life, we need to process many big data called network. Community is a very import property of network, because community used to explain the network, and make network useful. Now many community detecting algorithm can just detect disjointed community, they can not fit the network nowadays. For communities of network are usually overlapping,so overlapping detection algorithms will be more fit for network。We provide a new algorithm called Spectrum Based Clique Expansion(SCE). Based on finding all clique, SCE algorithm random walk on seeds, and output a local spectral space. In the local spectral space, mathematically, it is equivalent to solve linear programming problem. we solve the min 1-norm vector to get the community nearly. We do these operations in the subspace, so it is not related with size of graph. Because of independent working on seeds, so we can get overlapping community.As the result of SCE algorithm, SCE is very strong and stable to different seeds and community scoring function. Compared with other algorithms, SCE algorithm is better than GCE and DEMON algorithms. SCE algorithm can also apply to real data detecting.