A Novel Link Prediction Model in Multilayer Online Social Networks Using the Development of Katz Similarity Metric

The analysis of online social networks (OSNs) using graph theory is performed with the aim of extracting knowledge embedded in these networks. Link prediction (LP) problem is an important topic in the analysis of OSNs. LP refers to estimating the possibility of links between users in the future. LP in multilayer networks is the problem of finding links between users based on information from other layers. Similarity metrics between users is one of the most common techniques for solving LP. Nevertheless, the development of these metrics for multilayer networks has become an important challenge for researchers. This paper presents a multilayer OSN-based LP model through the analysis of Twitter and Foursquare networks. We propose a novel metric that calculates the similarity between users by considering the information of intralayer and interlayer links in a two-layer network. The proposed similarity metric applies topological features and reliable paths based on the Katz similarity metric. Here, we used the extracted topological features to map the network to a weighted network and reliable paths to consider the importance of each link in relationships. Experimental results show the effectiveness of the proposed similarity metric for LP in the single-layer and two-layer networks. Also, the comparisons show the superiority of the proposed metric compared to the classical similarity metrics, namely Katz and FriendLink, as well as equivalence algorithms such as Meta-Paths and SEM-Paths.