Life cycles of communities in an online social network

Abstract

We look into the community evolution in the Cyworld guestbook data. We use the hierarchical link clustering (HLC) approach to identify communities. HLC allows nodes to belong to multiple communities in contrast to other community identification algorithms that are node-based and allow no overlap. We define the following six key events in community evolution: birth, death, merge, split, growth, and shrink. We present our analysis results in terms of the six events. We then investigate how community evolves in the evolving networks. Our work reveals the life cycle of communities in online social network. Wefirst shows that the life expectancy of large communities is much longer than that of small communities, and large proportion of community members change over time. Then we find that the distributions of new-born and died communities are close. We discover that the larger communities split into a larger number of pieces and the pieces are similar sized. We show that higher relative node overlap leads to Merge of two communities.