How are You Affected? A Structural Graph Neural Network Model Predicting Individual Social Influence Status

Human’s daily life is now inextricably bound with online social networks like Weibo and Twitter, where user’s opinions and emotions are delivered and get influenced by each other frequently. Therefore, predicting and understanding such phenomenon of social influence, especially on individuals, becomes crucial for applications like recommendation and advertising. Although recent studies achieve some predictions on individual social influence status (active or inactive), the complexity and diversity of the social structures have not been well solved, and such works also lack a deep understanding of the influence mechanism itself like how individuals are affected. Therefore, a structural graph neural network model (SGN) is proposed to learn the diverse relationships and quantify the propagated influence between users. The SGN model is consists of two special representation modules and some global layers. The two representation modules are based on two major social network structures: friendship structure and influence propagation structure. In the modules, the well-developed graph neural networks, GCN (Graph Convolutional Network) and GAT (Graph Attention Networks), are respectively applied to capture spatial correlations. Besides, the global attention mechanism then helps to quantify the relationships between influencers and influencees. Finally, experiments on two real-world social networks show that the proposed SGN not only outperforms other baselines on prediction metrics, but is also able to reveal the intermediate neighbors who affect the target most.