Presenting a new motif-based link prediction for predicting activities in Facebook

Abstract

Motif distribution in different complex networks holds much information about network structure. As a network evolves, some new links appear and its motifs are transitioned into each other and their distribution changes as well as a network structure. The study of how motifs transition to each other can impact on study of complex networks and the link formation process. In this paper, the distribution and transitions of triads in different Facebook activity networks such as like, comment, post, and share networks are studied. After studying motif transitions over time, a new algorithm is presented for link prediction or activity recommendation for Facebook activity networks. In addition, to analyze motif transitions easily as well as speed up the presented algorithm, new concepts for analyzing sub-graphs have been presented including the motif transition graph and also its Hasse diagram. In addition, we have found out that among 53 different triad transitions only 10 of them are valuable in terms of link prediction in activity networks, and this has been used to accelerate the link prediction algorithm. The performed experiments show that the presented method has better results versus previous link prediction methods.