Knotenklassifikation
Node Classification ist eine maschinelles Lernen-Aufgabe in der Graphendatenanalyse, die darauf abzielt, unbekannte Attribute von Knoten in einem Graphen basierend auf den Attributen der Knoten und ihren Beziehungen vorherzusagen. Diese Aufgabe wird durch Modelle wie Graph Neural Networks (GNNs) umgesetzt, und ihre Leistung wird in der Regel anhand von Metriken wie Genauigkeit (Accuracy) und F1-Score auf Benchmark-Datensätzen wie Cora, Citeseer und Pubmed bewertet. Node Classification hat erheblichen Anwendungswert in Bereichen wie der Analyse sozialer Netzwerke, Empfehlungssystemen und Bioinformatik.
Cora
SSP
Citeseer
PathNet
Pubmed
ClusterGCN
Wisconsin
FSGNN (3-hop)
Actor
Texas
2-HiGCN
Chameleon
FSGNN (8-hop)
Cornell
Squirrel
CiteSeer with Public Split: fixed 20 nodes per class
Chameleon (60%/20%/20% random splits)
Film (60%/20%/20% random splits)
PubMed (60%/20%/20% random splits)
PubMed with Public Split: fixed 20 nodes per class
OGC
Squirrel (60%/20%/20% random splits)
Cora with Public Split: fixed 20 nodes per class
OGC
Cornell (60%/20%/20% random splits)
HH-GraphSAGE
Texas (60%/20%/20% random splits)
ACM-GCN++
Wisconsin (60%/20%/20% random splits)
ACM-GCN++
CiteSeer (60%/20%/20% random splits)
Cora (60%/20%/20% random splits)
Penn94
GloGNN++
Citeseer (48%/32%/20% fixed splits)
Geom-GCN
Cora (48%/32%/20% fixed splits)
NLGAT
genius
GESN
PubMed (48%/32%/20% fixed splits)
GCNII
Coauthor CS
CoLinkDist
PPI
g2-MLP
PascalVOC-SP
SAN+LapPE
COCO-SP
GPS
Reddit
BNS-GCN
Cora (0.5%)
CPF-ind_APPNP
Cora (1%)
VHCN
Cora (3%)
Truncated Krylov
AMZ Photo
JK (Heat Diffusion)
CiteSeer (0.5%)
MT-GCN
CiteSeer (1%)
Truncated Krylov
Coauthor Physics
CoLinkDist
PubMed (0.03%)
Truncated Krylov
PubMed (0.05%)
PubMed (0.1%)
Truncated Krylov
Amazon Computers
GAT
CLUSTER
GRIT
Amazon Photo
CoLinkDist
arXiv-year
LINKX
PATTERN
GRIT
Cora: fixed 20 node per class
Self-supervised GraphMAE
Cora Full-supervised
GCNII
PATTERN 100k
EGT
Yelp-Fraud
AM
R-GCN
Facebook
GNNMoE(GCN-like P)
Flickr
GCN+GAugM (Zhao et al., 2021)
AIFB
R-GCN
AMZ Comp
HH-GCN
BGS
SCENE
Brazil Air-Traffic
Citeseer Full-supervised
IncepGCN+DropEdge
Europe Air-Traffic
GCN_cheby (Kipf and Welling, 2017)
pokec
Polynormer
Pubmed Full-supervised
GraphSAGE+DropEdge
roman-empire
USA Air-Traffic
DEMO-Net(weight)
Amazon-Fraud
GTAN
Struc2vec
MUTAG
RR-GCN-PPV
Wiki-CS
CGT
Wiki-Vote
DEMO-Net(weight)
AMZ Computers
CPF-ind-GAT
Cora Full
CoLinkDist
DBLP
GRACE
Placenta
GraphSAGE
Wikipedia
GraphGAN
amazon-ratings
AVA
ASDNet [ASDNet_ICCV2021]
BlogCatalog
DEMO-Net(weight)
Chameleon (48%/32%/20% fixed splits)
GREAD-BS
Cora Full with Public Split
CoLinkDist
MAG-scholar-C
FastGCN
MAG240M-LSC
minesweeper
MuMiN-large
MuMiN-medium
MuMiN-small
LaBSE
NELL
DFNet-ATT
Squirrel (48%/32%/20% fixed splits)
tolokers
Polynormer
Cornell (48%/32%/20% fixed splits)
MS ACADEMIC
APPNP
AMPLUS
R-GCN
Chameleon(60%/20%/20% random splits)
Citeseer: fixed 20 node per class
Coauthor Phy
GCN-LPA
Cora: fixed 10 node per class
CPF-tra-GCNII
Cora: fixed 5 node per class
GraphMix (GCN)
Crocodile
LW-GCN
Deezer Romania
DMG777K
RR-GCN-PPV
DMGFULL
RR-GCN-PPV
MDGENRE
questions
GCN
Squirrel(60%/20%/20% random splits)
NCSAGE
Telegram (Directed Graph label rate 60%)
ScaleNet
Texas (48%/32%/20% fixed splits)
twitch-gamers
Dual-Net GNN
Wiki
DAOR
wiki
A2DUG
Wisconsin (48%/32%/20% fixed splits)
YouTube
LINE
20NEWS
GraRep
Amazon2M
Cluster-GCN
AMZ Computers: fixed 20 node per class
BGP
PathNet
Bitcoin-Alpha
GraphMix (GCN)
Bitcoin-OTC
GraphMix (GCN)
CellTypeGraph Benchmark
plantcelltype-EdgeDeeperGCN
Citeseer random partition
GraphMix (GCN)
CiteSeer with Public Split: fixed 5 nodes per class
GraphMix (GCN)
Cora: fixed 20 nodes per class
Cora random partition
GraphMix (GCN)
Deezer Croatia
Deezer Hungary
Electronics
Film(48%/32%/20% fixed splits)
NBA
Pubmed: fixed 20 node per class
Pubmed random partition
GraphMix (GCN)