DeepInf: Modeling influence locality in large social networks
Jiezhong Qiu,Jian Tang,Hao Ma,Yuxiao Dong,Kuansan Wang,Jie Tang
2018-01-01
Abstract:Social influence is everywhere around us, not only in our daily physical lives but also in the virtual Web space. The term social influence typically refers to the phenomenon that a person’s emotions, opinions, or behaviors are affected by others. With the emergence and rapid proliferation of online social platforms, people have seen the impact of social influence in every field, including but not limited to advertising, innovation adoption, presidential elections, stock markets, and civil unrest. There is little doubt that social influence has become a prevalent, subtle, and complex force that drives our social decisions and activities. Therefore, there is a clear need for methods and frameworks to characterize, understand, and quantify the underlying mechanism and dynamics of social influence. Indeed, extensive work has been done on social influence prediction in the literature [10, 23, 27]. For example,[10] studied predicting the sizes of cascades with different kinds of hand-crafted features extracted from networks while [23] proposed an end-toend approach for the problem. All these approaches mainly study predicting the global patterns of social influence, ie, macro-level social influence prediction. However, in many online applications such as advertising and recommendation, how to effectively predict the social influence for each user is critical, ie, micro-level social influence prediction.In this paper, we focus on micro-level social influence prediction. We formulate it as a classification problem based on the definition of influence locality inherited from Zhang et al.[50]. Under this formulation, we want to predict the action status of a user given