Incorporating communities’ structures in predictions of missing links

Abstract

This article introduces a community-based approach to link prediction that identifies the links likely to be seen in the near future in a network. The proposed method incorporates community structure as a feature in the predictions of missing links in a network. We design a feature-based similarity measure that considers the impact of community structure in addition to other network features in link prediction. We analyze the performance of the devised approach in terms of precision, recall, accuracy, and area-under-the-curve (AUC) metrics on real-world datasets. Further, we examine the performance of the devised method in terms of execution time against real-world and synthetic datasets. The proposed approach outperforms the other existing approaches, as will be shown experimentally later.