Implementing link-prediction for social networks in a database system

Storing and querying large social networks is a challenging problem, due both to the scale of the data, and to intricate querying requirements. One common type of query over a social network is link prediction, which is used to suggest new friends for existing nodes in the network. There is no gold standard metric for predicting new links. However, past work has been effective at identifying a number of metrics that work well for this problem. These metrics vastly differ one from another in their computational complexity, e.g., they may consider a small neighborhood of a node for which new links should be predicted, or they may perform random walks over the entire social network graph. This paper considers the problem of implementing metrics for link prediction in a social network over different types of database systems. We consider the use of a relational database, a key-value store and a graph database. We show the type of database system affects the ease in which link prediction may be performed. Our results are empirically validated by extensive experimentation over real social networks of varying sizes.