Deep attributed network representation learning via attribute enhanced neighborhood

Attributed network representation learning aims at node embeddings by integrating structure and attribute information. It is a challenge to fully capture the microscopic semantics simultaneously, where includes one-step, two-step multi-step relations, indicating first-order, second-order high-order proximity of nodes, respectively. In this paper, we propose deep attributed via enhanced neighborhood (DANRL-ANE) model improve robustness effectiveness representations. The DANRL-ANE adopts idea autoencoder, expands decoder component three branches different order proximity. We linearly combine adjacency matrix with similarity as input DANRL-ANE, calculated cosine between attributes based on social homophily. Moreover, sigmoid cross-entropy loss function extended character, so that first-order could be well-preserved. compare our state-of-the-art models, especially, latest graph autoencoders (GAEs) method, ARGA, demonstrate contribution each module real-world datasets two analysis tasks, i.e., link prediction classification. performs well various networks, even sparse networks or isolated nodes when information sufficient.