Edge but not Least: Cross-View Graph Pooling
- Publisher:
- Springer Nature
- Publication Type:
- Conference Proceeding
- Citation:
- Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2022, 13714 LNAI, pp. 344-359
- Issue Date:
- 2022-01-01
Closed Access
| Filename | Description | Size | |||
|---|---|---|---|---|---|
| 978-3-031-26390-3.pdf | Published version | 44.63 MB |
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
Graph neural networks have emerged as a powerful representation learning model for undertaking various graph prediction tasks. Various graph pooling methods have been developed to coarsen an input graph into a succinct graph-level representation through aggregating node embeddings obtained via graph convolution. However, because most graph pooling methods are heavily node-centric, they fail to fully leverage the crucial information contained in graph structure. This paper presents a cross-view graph pooling method (Co-Pooling) that explicitly exploits crucial graph substructures for learning graph representations. Co-Pooling is designed to fuse the pooled representations from both node view and edge view. Through cross-view interaction, edge-view pooling and node-view pooling mutually reinforce each other to learn informative graph representations. Extensive experiments on one synthetic and 15 real-world graph datasets validate the effectiveness of our Co-Pooling method. Our results and analysis show that (1) our method is able to yield promising results over graphs with various types of node attributes, and (2) our method can achieve superior performance over state-of-the-art pooling methods on graph classification and regression tasks.
Please use this identifier to cite or link to this item: