A review on predicting evolution of communities

Abstract

In recent years, research on dynamic networks has increased as the availability of data has grown tremendously. Understanding the dynamic behavior of networks can be studied at the mezzo-scale (e.g., at the community level), as communities are the most informative structure in nonrandom networks and also evolve over time. Tracking the evolution of communities can provide evolution patterns to predict their future development. For example, a community may either grow into a larger community, remain stable, shrink into a smaller community, split into several smaller communities, or merge with another community. Predicting these evolutions is one of the most difficult problems in social networks. Better predictions of community evolution can provide useful information for decision support systems, especially for group-level tasks. So far, this problem has been studied by some researchers. However, there is a lack of a survey/review of existing work. This has prompted us to conduct this study. In this paper, we first categorize the existing works according to their methodological principles. Then, we focus on the works that use machine learning classifiers for prediction in this decade as they are in majority. We then highlight open problems for future research. In this way, this paper provides an up-to-date overview and a quick start for researchers and developers in the field of community evolution prediction.