Communities Found by Users -- not Algorithms

Many algorithms have been created to automatically detect community structures in social networks. These algorithms have been studied from the perspective of optimisation extensively. However, which community finding algorithm most closely matches the human notion of communities? In this paper, we conduct a user study to address this question. In our experiment, users collected their own Facebook network and manually annotated it, indicating their social communities. Given this annotation, we run state-of-the-art community finding algorithms on the network and use Normalised Mutual Information (NMI) to compare annotated communities with automatically detected ones. Our results show that the Infomap algorithm has the greatest similarity to user defined communities, with Girvan-Newman and Louvain algorithms also performing well.