.png){kind=icon}
.png){kind=icon}
Return
Capturing local and global information: Multi-view graph convolutional network via granular-ball computing and collaborative matrix
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.129057.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In the field of multi-view learning based on graph convolutional network (GCN), many studies focus on integrating information from different views to enhance model performance. Most of them primarily extract consistency information through feature fusion or topology fusion. However, this fusion mode overlooks local information within individual views, which inherently limits the model performance. To overcome this limitation, this paper proposes the multi-view Graph Convolutional Network via Granular-ball computing and Collaborative Matrix (GBCM-GCN), which consists of two modules: the granular-ball based topology construction module and collaborative matrix based convolution module. The former utilizes the boundary distance between granular-balls to construct high-quality topology based on both local and global connections. The latter leverages convolution kernels of different dimensions to extract local and global information of embedding representation during the convolution process, then utilizes learnable parameters to integrate them into a collaborative matrix. Furthermore, the collaborative matrix is shared across all views, which can enhance the consistency representation across different views. Finally, the experimental result show that GBCM-GCN outperforms existing multi-view semi-supervised classification methods.
Keywords:
multi-view learning
graph convolutional network
granular-ball computing
collaborative matrix
semi-supervised classification
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W