.png){kind=icon}
.png){kind=icon}
Return
AEGraph: Node attribute-enhanced graph encoder method
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2023.121382.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Graph representation learning faces challenges in node classification tasks due to the intricate interplay between node categories, topological structure, significant characteristics, node attributes, and label information. How-ever, prevailing graph convolutional networks (GCNs) exhibit inefficiencies in information propagation and need more robustness when dealing with missing node attributes. To surmount these issues, attribute-based dynamic wandering and attribute enhancement methods are proposed. First, the Node Attribute Random Wandering (NARW) method is proposed to obtain the node sequences. The NARW method fully considers the node-to-node attribute information to determine the wandering direction of the sequences. Subsequently, the node attribute enhanced graph encoder model (AEGraph) is proposed, which aims to improve the node classification perfor-mance by combining the graph structure and node attribute information during the information propagation process. Each iteration process includes node label information aggregation and node label information update. Finally, to fully consider the local and global information of the graph, this paper reduces the noise generated by information propagation during the iteration process. The model proposed in this paper shows promising results with a low labeling rate. Experimental results show that the method achieves better results and significant improvements on different datasets.
Keywords:
Node classification
Semi -supervised learning
Graph convolutional networks
Node attribute augmented
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
Organization
No organization information available