DeepInf

Abstract

Social and information networking activities such as on Facebook, Twitter,\nWeChat, and Weibo have become an indispensable part of our everyday life, where\nwe can easily access friends' behaviors and are in turn influenced by them.\nConsequently, an effective social influence prediction for each user is\ncritical for a variety of applications such as online recommendation and\nadvertising.\n Conventional social influence prediction approaches typically design various\nhand-crafted rules to extract user- and network-specific features. However,\ntheir effectiveness heavily relies on the knowledge of domain experts. As a\nresult, it is usually difficult to generalize them into different domains.\nInspired by the recent success of deep neural networks in a wide range of\ncomputing applications, we design an end-to-end framework, DeepInf, to learn\nusers' latent feature representation for predicting social influence. In\ngeneral, DeepInf takes a user's local network as the input to a graph neural\nnetwork for learning her latent social representation. We design strategies to\nincorporate both network structures and user-specific features into\nconvolutional neural and attention networks. Extensive experiments on Open\nAcademic Graph, Twitter, Weibo, and Digg, representing different types of\nsocial and information networks, demonstrate that the proposed end-to-end\nmodel, DeepInf, significantly outperforms traditional feature engineering-based\napproaches, suggesting the effectiveness of representation learning for social\napplications.\n