Adaptive Graph Encoder for Attributed Graph Embedding
Experimental results show that AGE consistently outperforms state-of-the-art graph embedding methods considerably on node clustering and link prediction tasks, and the proposed Adaptive Graph Encoder employs an adaptive encoder that iteratively strengthens the filtered features for better node embeddings.
Abstract
Attributed graph embedding, which learns vector representations from graph\ntopology and node features, is a challenging task for graph analysis. Recently,\nmethods based on graph convolutional networks (GCNs) have made great progress\non this task. However,existing GCN-based methods have three major drawbacks.\nFirstly,our experiments indicate that the entanglement of graph convolutional\nfilters and weight matrices will harm both the performance and robustness.\nSecondly, we show that graph convolutional filters in these methods reveal to\nbe special cases of generalized Laplacian smoothing filters, but they do not\npreserve optimal low-pass characteristics. Finally, the training objectives of\nexisting algorithms are usually recovering the adjacency matrix or feature\nmatrix, which are not always consistent with real-world applications. To\naddress these issues, we propose Adaptive Graph Encoder (AGE), a novel\nattributed graph embedding framework. AGE consists of two modules: (1) To\nbetter alleviate the high-frequency noises in the node features, AGE first\napplies a carefully-designed Laplacian smoothing filter. (2) AGE employs an\nadaptive encoder that iteratively strengthens the filtered features for better\nnode embeddings. We conduct experiments using four public benchmark datasets to\nvalidate AGE on node clustering and link prediction tasks. Experimental results\nshow that AGE consistently outperforms state-of-the-art graph embedding methods\nconsiderably on these tasks.\n