We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Multi-View Block Matrix-Based Graph Convolutional Network.
- Authors
Kaibiao Lin; Runze Chen; Jinpo Chen; Ping Lu; Fan Yang
- Abstract
In recent years, graph convolutional networks (GCNs) have gained significant attention for graph embedding learning. However, the efficacy of GCNs and their variants is often constrained by the implicit homophily assumption, which presumes nodes in close proximity to exhibit similar features. This assumption becomes particularly limiting in the context of complex heterogeneous graphs, as it restricts GCNs' ability to aggregate diverse node information. Addressing this, we propose a novel aggregation mechanism adept at discerning homophilic from heterophilic neighborhoods, thus achieving effective "classification aggregation". More specifically, this study combines block modeling with the aggregation process, allowing the GCNs to learn the aggregation rules across various classes of neighbors automatically. On this basis, a multi-view module is designed to extract semantic information from each view, and the final embedding is obtained by aggregating the information from all views through the attention module. Experiments indicated that the proposed method performs optimally across both heterogeneous and homogeneous graph datasets.
- Subjects
GRAPH neural networks
- Publication
Engineering Letters, 2024, Vol 32, Issue 6, p1073
- ISSN
1816-093X
- Publication type
Article