School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing, China 100083; Beijing Key Laboratory of Knowledge Engineering for Materials Science, Beijing, China100083

---

Abstract—Since the conventional algorithm for community structure detection in a stand-alone environment cannot handle the giant network whose number of nodes is more than 105, and the widely used MapReduce method has a limitation on dealing with excessive I/O operations during the iterative process, an efficient parallel computing method based on BSP (Bulk Synchronous Parallel) model for detecting community structure is proposed in this paper. The Fast Newman method is improved into parallel calculations with multiple steps under the framework of BSP model. It is more efficient to discover community structures in the large scale network. In order to testify the performance of the proposed method, a hama platform was built up on the same cluster of the hadoop platform. And a dataset, at a scale of 106, was also simulated for the experiments. It is approved that the proposed method is not only able to solve the issue of memory overrun in the conventional calculation on a stand-alone computer, but also to improve the performance effectively comparing to the MapReduce model. The proposed method has high practical value in large scale networks.

Index Terms—complex networks, graph clustering, modularity, Fast-Newman algorithm, BSP model

[PDF]

Cite: Yi Sun, Zhen Hua, Li-hui Zou, "A Parallel Computing Method for Community Structure Detection Based on BSP Model," Journal of Software vol. 9, no. 7, pp. 1876-1885, 2014.