Mining of Network Communities by Spectral Characterization Using KD-tree

Abstract

Different methods and algorithms have been employed to carry out the task of community mining. Conversely, in the real world, many applications entail distributed and dynamically evolving networks, wherein resources and controls are not only decentralized, but also restructured frequently. This leads a problem of finding all communities from a given network, within that the links are dense, but between which they are sparse. This is referred as Network Community Mining Problem (NCMP). A network community contains a group of nodes connected based on certain relationships sometimes that refers to a special sort of network arrangement where the community mining is discovering all communities hidden in distributed networks based on their relevant local outlooks. To avoid the above mentioned problem, the existing work presented a novel model for characterizing network communities via introducing a stochastic process on networks and analyzing its dynamics based on the large deviation theory. By Using the fundamental properties of local mixing, then proposed an efficient implementation for that framework, called the LM (Network community mining based on Local Mixing properties) algorithm. There also some drawbacks are identified. Actual number of communities is estimated by using a recursive bisection approach, jointly with a predefined stopping criterion. The recursive bisection strategy does not optimize communication performance and the complexity of performing the partitioning. To solve these problems in proposed work a new community bipartition scheme is developed by using KD-Tree. Also, the stopping criterion is calculated automatically by efficiently determining the minimum Eigen-gap without explicitly computing eigenvalues. The experimental result shows that the proposed scheme is more effective and scalable when compared with the existing scheme.