Neoknn: A Network Embedding Method Only Knowing Neighbor Nodes

Recently, network embedding has attracted increasing research attention, due to its effectiveness in a variety of applications. Most of the existing work aims to embed varied kinds of networks, such as plain networks, dynamic networks, and attributed networks. However, all of them require sufficient information such as network structure or node attributes. In practice, we can hardly obtain the entire network structure and abundant attributes. The network information provided is often limited. For example, we may only know a few neighbors of each node. In this paper, we address an important embedding problem (i.e., embedding nodes in a limited situation), and propose a novel deep network embedding framework in a soft coupling way to infer node embedding in such a situation. Extensive experimental results demonstrate that, given limited information, our representation significantly outperforms the representations learned by state-of-the-art methods.