A classification approach to link prediction in multiplex online ego-social networks

Online social networks are the inseparable element of current modern societies and significantly influence forming and consolidating social relationships. In nature, these networks are multiplex so that multiple links may exist between the same two users across different social networks. In this paper, we study the ego-social features of multiplex links, spanning more than one social networks and apply their structural and interaction features to the problem of link prediction. The link prediction is applied for various cases in social networks such as new recommendations for users, friendship suggestions and fake relations discovery. Most of the real-world social networks promote communications in multi-layers (for example, the platform of multiple social networks). In this work, the problem of link prediction in multiple networks including Twitter (as a microblogging service) and Foursquare (as a place-based social network) has been studied. We consider the users jointly use both social network platforms and develop a classification algorithm for predicting the links. Hereto, the layers structural information is considered to predict the links in Foursquare network. Technically, solving this classification problem is accomplished through defining three sets of features based on nodal structure, ego-paths and meta-paths (SEM-Path). Three classic classifiers such as ID3, SVM and LR are used for the classification problem in the SEM-Path method. Our evaluations show that we can successfully predict links across social networking platforms. In fact, evaluations aim to shed light on the implications of multiplexity for the link generation process. The SVM classifier outperforms other classifiers with an average precision equal to 77.62%. Also, it has almost 1.5% superiority than the meta-path-based algorithm method.