Link Prediction across Heterogeneous Social Networks: A Survey

Online social networks have gained great success in recent years. Some online social networks only involving users and social links among users can be represented as homogeneous networks. Meanwhile, some other social networks containing abundant information, which include multiple kinds of nodes and complex relationships, can be denoted as heterogeneous networks. Predicting the missing links or links that will be formed in the future based on a snapshot of social networks is formally defined as the link prediction problem. Link prediction problems have extensive applications in real-world social networks and many concrete social services can be cast as link prediction tasks, e.g., friend and location recommendations can all be solved as the problem of predicting social links among users and the location links between users and locations. Link prediction problems have been an important research topic for many years and a large number of different methods have been proposed so far. This article summaries the existing link prediction methods for both homogeneous and heterogeneous networks, which include various unsupervised link predicators, random walk based link prediction methods, methods based on matrix factorization techniques, supervised link prediction methods and meta paths based link prediction methods. Meanwhile, as proposed in recent works, people are usually involved in multiple social networks simultaneously nowadays and networks sharing common users are formally defined as the aligned networks. In this article, we will also introduce the latest progress of link prediction problems across multiple aligned heterogeneous networks. The link prediction problems across aligned networks can include anchor link prediction problem and link transfer across aligned heterogeneous networks. We will introduce the newly proposed methods to solve these problems in details and, finally, we will conclude this survey with a discussion about the future link prediction research works.