A parallel community detection algorithm based on spanning trees

The analysis of large and complex graph structures such as the web graph has spawned a plethora of computationally challenging problems. Community detection is perhaps one of the most important among those. The algorithms of GirvanNewman and Louvain are two early, well-known methods to attack the problem. Nevertheless, many interesting approaches have since then been presented in the literature. One of these efforts, called ST algorithm, involves the use of spanning tree computations and the neighborhood overlap metric. As the size of input graphs grows rapidly, parallel algorithms provide a natural way to handle the very large amount of data. In this work we propose a parallel version of the ST algorithm, since all its components admit fast parallelization. We argue that the parallel ST scales well and that it can lend itself to efficient implementations in parallel computing environments.