Aligning Users Across Social Networks via Intra and Inter Attentions

In recent years, aligning users across different social networks receives a significant attention. Previous studies solve the problem based on attributes or topology structure approximation. However, most of them suffer from error propagation or the noise from diverse neighbors. To address the drawback, we design intra and inter attention mechanisms to model the influence of neighbors in local and across networks. In addition, to effectively incorporate the topology structure information, we leverage neighbors from labeled pairs instead of these in original networks, which are termed as matched neighbors. Then we treat the user alignment problem as a classification task and predict it upon a deep neural network. We conduct extensive experiments on six real-world datasets, and the results demonstrate the superiority of the proposed method against state-of-the-art competitors.