Performance Comparison for Different Approaches using Micro-Averaged F1-Score

Approach	Transductive Supervised	Transductive Unsupervised	Inductive Supervised	Inductive Unsupervised
Random Baseline	0.20	0.20	0.20	0.20
Raw Features	0.59	0.59	0.49	0.49
GCN Kipf et al. (2017)	0.58	0.60	0.43	0.61
GraphSAGE-MEAN Hamilton et al. (2017)	0.62	0.67	0.60	0.60
GraphSAGE-MEANPOOL Hamilton et al. (2017)	0.81	0.69	0.45	0.61
GraphSAGE-MAXPOOL Hamilton et al. (2017)	0.80	0.68	0.44	0.55
GraphSAGE-LSTM Hamilton et al. (2017)	0.81	0.69	0.45	0.59
GAT Velickovic et al. (2018)	0.75	0.69	0.43	0.51
GIN Xu et al. (2018)	0.78	0.69	0.46	0.47
GAIN (ours)	0.81	0.71	0.59	0.56
% gain over Baseline	37%	20%	22%	24%