Interlayer link prediction based on multiple network structural attributes

Abstract

Information may flow across multiple social network platforms and spread simultaneously on them. These platforms constitute a multiplex network in which each social network platform functions as a subnetwork. One problem in information propagation analysis on the multiplex network is to reveal the corresponding relationship of accounts belonging to the same user across different subnetworks. It can be summarized as the interlayer link prediction problem in a multiplex network. To predict more unobserved interlayer links accurately, most of the current structure-based approaches adopt an iterative strategy. However, iterative manners suffer from a big challenge of high time consumption. In this paper, we propose an interlayer link prediction framework based on multiple structural attributes (MulAtt). It calculates the matching degree of unmatched nodes once by leveraging the information of closed triad, intralayer links, matched neighbors and intralayer links of neighbors simultaneously to guarantee accuracy while reducing time consumption. We also propose a fast possible matched closed triad counting algorithm to improve the efficiency of obtaining the clues from the closed triad. Experiments on four widely used real-world multiplex networks demonstrate that the MulAtt framework can achieve better performance than several existing network structure-based methods in a non-iterative way.