A synchronous feature learning method for multiplex network embedding.

Compared with single-layer networks, multiplex networks can describe real-world scenarios in more detail while suffering from requiring considerable computing and storage resources at the same time. Network feature learning, which aims to embed networks into a low dimensional space, is an effective method for solving these problems. Currently, research on multiplex network embedding faces two major challenges: how to make full use of the connected information in different layers and how to embed multiplex networks into a unified space. In this paper, a novel multiplex network embedding model is proposed to solve these two problems. It preserves all the first-, second- and multi-order proximities in multiplex networks by optimizing the corresponding objective functions. The network reconstruction step combines information of different types of relations in other layers while maintaining their distinctive properties. The proposed synchronous learning strategy provides a path to embed multiplex networks into a unified space. Extensive experiments on three real applications: visualization, link prediction and node classification are conducted to validate the effectiveness of the proposed method. The experimental results show that it achieves better or comparable performance compared with several state-of-the-art methods. [ABSTRACT FROM AUTHOR]