PNE: Label embedding enhanced network embedding

Unsupervised NRL (Network Representation Learning) methods only consider the network structure information, which makes their learned node representations less discriminative. To utilize the label information of the partially labeled network, several semi-supervised NRL methods are proposed. The key idea of these methods is to merge the representation learning step and the classifier training step together. However, it is not flexible enough and their parameters are often hard to tune. In this paper, we provide a new point of view for semi-supervised NRL and present a novel model named Predictive Network Embedding (PNE). Briefly, we embed nodes and labels into the same latent space instead of training a classifier in the representation learning process. Thus the discriminability of node representations is enhanced by incorporating the label information. We conduct node classification task on four real world datasets. The experimental results demonstrate that our model significantly outperforms the state-of-the-art baselines.