Enhancing Signed Graph Attention Network by Graph Characteristics: An Analysis
Issued Date
2024-05-01
Resource Type
ISSN
22178309
eISSN
22178333
Scopus ID
2-s2.0-85196088718
Journal Title
TEM Journal
Volume
13
Issue
2
Start Page
885
End Page
896
Rights Holder(s)
SCOPUS
Bibliographic Citation
TEM Journal Vol.13 No.2 (2024) , 885-896
Suggested Citation
Kaewhit P., Lewchalermvongs C., Lewchalermvongs P. Enhancing Signed Graph Attention Network by Graph Characteristics: An Analysis. TEM Journal Vol.13 No.2 (2024) , 885-896. 896. doi:10.18421/TEM132-05 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/98914
Title
Enhancing Signed Graph Attention Network by Graph Characteristics: An Analysis
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
– A graph neural network (GNN) is one of successful methods for handling tasks on a graph data structure, e.g. node embedding, link prediction and node classification. GNNs focus on a graph data structure that must aggregate messages on nodes in the graph to retain a graph-structured information in new node’s message and proceed tasks on a graph. One of modifications on the propagation step in GNNs by adopting attention mechanism is a graph attention network (GAT). Applying this modification to signed graphs generated by sociological theories is called signed graph attention network (SiGAT). In this research, we utilize SiGAT and create novel graphs using graph characters to assess the performance of SiGAT models embedded in nodes across various characteristic graphs. The primary focus of our study was linked prediction, which aligns with the task employed in the previous research on SiGAT. We propose a method using graph characteristics to improve the time spent on the learning process in SiGAT.