Random Walk Graph Auto-Encoders With Ensemble Networks in Graph Embedding

Recently graph auto-encoders have received increasingly widespread attention as one of the important models in field deep learning. Existing auto-encoder only use convolutional neural networks (GCNs) encoders to learn embedding representation nodes. However, GCNs are suitable for transductive learning, poor scalability and shallow with a perceptual field, limitations node feature extraction. To alleviate these problems, we propose an adaptive weight integration network (GAT) GCN’s random walk (EGRWR-GAE) better There is large amount noise data, which interferes extraction GAT model sensitive noisy (EGSRWR-GAE) that integrates GAT, GCN, self-supervised (SuperGAT) using weights. The effectiveness our well demonstrated by three publicly available datasets (Cora, Citeseer, Pubmed) optimizations up 2.2% on link prediction task 12.9% clustering task.