Empirical evaluation of applying ensemble methods to ego-centred community identification in complex networks

Unfolding the community structure of complex networks is still to be one of the most important tasks in the field of complex network analysis. However, in many real settings, we seek to uncover the community of a given node rather than partitioning the whole graph into communities. A main trend in the area of ego-centred community identification consists in applying a greedy optimization approach of a local modularity measure. Different local modularity functions have been proposed in the scientific literature. In this work, we explore applying different ensemble approaches in order to combine different local modularity functions. Explored approaches include naive combine-and-rank approach, ensemble ranking approaches, and ensemble clustering. Experiments are conducted on different real and artificial benchmark networks for which ground truth community partitions are known. Results show that ensemble-ranking approaches provide better results than both state-of-the art approaches and other ensemble approaches.