Attribute Network Alignment Based on Network Embedding

Nodes with similar network structure and attribute features probably distribute across different networks. For instance, people tend to have accounts across various social networks. In recent years, network alignment to identify potential correspondences between nodes across networks has been research focus on social computing. In this paper, we propose an attribute network alignment method ANANE based on network embedding, which uses the network structure and node attributes together. Different from the previous embedding method based only on network structure and the existing iterative process based on structure and attributes, the proposed ANANE integrates heterogeneous network structure and attribute features into a unified embedding for node similarity measurement. We solve both the attribute network embedding and the network alignment simultaneously under a unified framework. In particular, we use neighbor approximation to generate the structure embedding and an auto-coder to obtain the attribute embedding. Then the attention mechanism is used to get the unified embedding for alignment. Empirically, we evaluate our proposed model ANANE over several real-world datasets, and it demonstrates effectiveness compared with several state-of-the-art methods on network alignment tasks.