Node Classification on Non-Homophilic (Heterophilic) Graphs
The task of node classification on non-homophilous graphs (heterophilous graphs) aims to evaluate the performance of models specifically designed for heterogeneous datasets. This task focuses on graphs where edges between different classes are more common than edges within the same class, and through systematic testing and analysis, it reveals differences in how models perform when dealing with heterophilous graphs, providing crucial references for optimizing graph neural networks.
Cornell (60%/20%/20% random splits)
ACMII-GCN
Wisconsin(60%/20%/20% random splits)
ACM-GCN++
Texas(60%/20%/20% random splits)
Chameleon(60%/20%/20% random splits)
ACM-GCN+
Chameleon (48%/32%/20% fixed splits)
Squirrel (48%/32%/20% fixed splits)
Penn94
Deezer-Europe
ACMII-GCN+++
Cornell (48%/32%/20% fixed splits)
Film(48%/32%/20% fixed splits)
genius
ClenshawGCN
twitch-gamers
Wisconsin (48%/32%/20% fixed splits)
O(d)-NSD
Texas (48%/32%/20% fixed splits)
Pubmed