Inferring Social Relationships across Social Networks for Viral Marketing

Abstract

Node classification in social networks is an important problem that has been widely studied in recent years. Several existing node classification methods mainly focus on identifying node classes by exploiting structural and attribute information. However, the information in an emerging information network is usually limited. For example, an emerging social networking service usually has very few registered users (referred to as active users) and a significant amount of new comers (referred to as non-active users) resulting in very sparse interactions among active users. Under this circumstances, distinguishing the users that is likely to be an active user in the future from large-scale new comers becomes challenging. In this paper, we propose a hybrid classification model, which can distinguish whether a non-active user will become an active user in the future by incorporating multiple relations through a unified ranking measure. More specifically, given a friendship network and a mobile communication network, we aim to discover a ranked list of users, who are likely to become active users in the future, from a massive amount of non-active users. We reported several empirical observations from real data sets and conducted extensive experiments to demonstrate the effectiveness of our hybrid classification model and ranking strategy.