DSANE:A Dual Structure-Aware Network Embedding Approach for User Identity Linkage

Increasing popularity and diversity of social media networks has encouraged people to participate on multiple online social networks. Thus, linking account across different networks belonging to the same person becomes an emerging task and is important to many downstream tasks. The mainstream approach is to regard users as nodes in networks and embed them in a low-dimensional latent space, making it convenient to input into subsequent predictive models. Existing methods are still faced up with challenges of incompleteness of networks and insufficiency performance. We propose a novel method called DSANE-Dual Structure-Aware Network Embedding for user identity linkage based on deep neural network, which firstly optimizes networks by a complete-filter method, and then embeds both local and global information into nodes’ vectors, thus greatly enriches nodes’ representation. Experiments on real-world datasets show that our method greatly outperforms baselines.