This work deals with the study of Variational Autoencoder framework in the context of homophilic graph data and its utility in network embedding. We first propose two model-based architectures to enhance the modeling and generative capacity of Variational Graph Autoencoder (VGAE) by mitigating the effect of over-pruning and explicitly incorporating the larger neighborhood. Afterward, we zoom in on a sub-problem of network embedding, i.e., community-aware network embedding for both homogeneous and heterogeneous graphs. For all the methodologies, we report extensive comparisons with direct SOTA competitors.
«
This work deals with the study of Variational Autoencoder framework in the context of homophilic graph data and its utility in network embedding. We first propose two model-based architectures to enhance the modeling and generative capacity of Variational Graph Autoencoder (VGAE) by mitigating the effect of over-pruning and explicitly incorporating the larger neighborhood. Afterward, we zoom in on a sub-problem of network embedding, i.e., community-aware network embedding for both homogeneous an...
»