Follow Recommendation in Social Networks.

Social Networks (SN) constitute a major component of our daily lives. Not only do they allow users to stay in touch with friends and family, but they also allow them to follow their favorite celebrities or people of interest. Such relationships within the social graph are most of the time one-directional and SN where this type of relations exists are directed networks. Link Prediction (LP) is used to analyze these networks and predict the creation of links in the future. However, most of LP metrics that have been proposed consider only undirected graphs, and leave out the most relevant aspect of a directed network which is the asymmetrical nature of its edges. In this work, we review proposed adaptations of LP measures for directed graphs and propose our own adaptation of a novel LP metric, which takes into account path depth and in-degrees and out-degrees of nodes. Using Area Under Curve (AUC), we compare the accuracy of all these measures on 3 directed social networks.