Deep Multi-Attributed-View Graph Representation Learning

Graph representation learning aims at mapping a graph into lower-dimensional feature space. Deep attributed representation, utilizing deep models on the structure and attributes, shows its significance in mining complex relational data. Most existing assume attributes single-attributed view. However, rich information real-world applications demands ability to handle multiple views. For example, social network users’ profiles posts represent two distinct A view or simple ensemble of them fails relations therein. To confront this challenge, paper proposes novel unsupervised model, called Multi-attributed-view Convolutional AutoEncoder (MagCAE). MagCAE learns node-pairwise proximity among multi-attributed views node embeddings, across which loss function is designed preserve likelihood. An aggregation layer specially developed optimize weights embeddings The extensive experiments four datasets demonstrate superiority over twelve baselines.