.

ISSN 2063-5346
For urgent queries please contact : +918130348310

Diffusion-based graph representation learning algorithms: Graph Diffusion Network (GDN), Graph Convolutional Networks (GCN) and Graph Diffusion System (GDS)

Main Article Content

B.Malathi, Dr.S.Chelliah
» doi: 10.48047/ecb/2023.12.si4.342

Abstract

Graph representation learning has emerged as a crucial task in the field of machine learning due to the widespread use of graphs in various domains such as social networks, biological networks, and citation networks. The primary goal of graph representation learning is to learn low-dimensional representations of graph nodes that can capture the underlying structure of the graph and be used for downstream tasks such as node classification, link prediction, and graph clustering.However, traditional graph representation learning methods rely on the assumption of local smoothness, which may not be sufficient for dealing with large, complex graphs where the local structure may not fully capture the global structure. To overcome this limitation, advanced methods such as the Graph Diffusion Network (GDN) have been proposed to capture both local and global information.GDN is designed to maintain both local and global consistency of the graph, and it does so by using a graph diffusion system to control the random walk of information flow and sense high-order local relationships in the graph. This approach allows GDN to capture both local and global information and learn intrinsic node representations in a progressive manner.One of the key advantages of GDN is its ability to self-refine on the graph structure, which enables it to learn the intrinsic node representations in a more efficient and effective manner compared to traditional methods that require the entire graph to be processed at once. Experiments on node classification tasks demonstrate that GDN outperforms traditional methods and achieves state-of-the-art results on several benchmark datasets.Overall, the success of GDN highlights the importance of capturing both local and global information in complex graphs and has the potential to advance graph representation learning in various domains

Article Details