Local Oriented Efficient Detection of Overlapping Communities in Large Networks

Overlapping community detecting for large-scale social networks becomes a research focus with the development of online social network applications. Among the current overlapping community discovery algorithms, LFM is based on local optimization of a fitness function, which is in consistent with the local nature of community, especially in large networks. But the original LFM may fall in loops when finding community memberships for some overlapping nodes and consumes still too much time when applied in large-scale social networks with power-law community size distribution. By limiting each node to be a seed at most once, LFM can avoid loop but fail to assign community memberships to some overlapping nodes. Based on the structural analysis, we found that the loop is due to the dysfunction of the fitness metric as well as the random seed selection used in LFM. To improve the detecting quality and computation efficiency of LFM, we propose a local orientation scheme based on clustering coefficient and several efficiency enhancing schemes. With these schemes, we design a modified algorithm LOFO (local oriented fitness optimization). Comparison over several large-scale social networks shows that LOFO significantly outperforms LFM in computation efficiency and community detection goodness.