200

Train

• 10,000 random generate graph
• trainset
• networkx.generators.random_graphs.powerlaw_cluster_graph(n=200, m=4, p=0.05)

Valid

• 100 random generate graph
• validset
• networkx.generators.random_graphs.powerlaw_cluster_graph(n=200, m=4, p=0.05)

Different Settings

GT/PR	Sigmoid(y_pr)	1/0	y_pr	BC_hat
Sigmoid(y_gt)	1. BCE Loss
	3. MSE Loss
1/0	2. BCE Loss	6. BCE Loss
y_gt			4. MSE Loss
BC				5. MSE Loss

Train & Validation

Train

Validation

Test Result

• CPU

Synthetic 5000

• data link

Method	top-1%	top-5%	top-10%	Kendal	Time(s)
1	0.93±0.03	0.89±0.02	0.87±0.02	0.70±0.01	0.34±0.01
2	0.88±0.03	0.90±0.02	0.87±0.02	0.71±0.01	0.38±0.11
3	0.93±0.03	0.88±0.02	0.85±0.02	0.69±0.01	0.36±0.02
4	0.93±0.02	0.87±0.02	0.84±0.02	0.65±0.02	0.41±0.03
5	0.93±0.03	0.89±0.02	0.87±0.02	0.68±0.01	0.38±0.03
6	0.00±0.00	0.03±0.01	0.25±0.02	0.19±0.01	0.37±0.02

youtube

• data link

Method	top-1%	top-5%	top-10%	Kendal	Time(s)
1	0.63	0.54	0.60	0.31	94.24
2	0.63	0.69	0.71	0.48	88.69
3	0.59	0.49	0.54	0.15	89.21
4	0.63	0.58	0.52	0.23	92.11
5	0.62	0.45	0.45	0.24	87.17
6	0.00	0.05	0.09	0.29	88.32

Amazon

Scale	top-1%	top-5%	top-10%	Kendal	Time(s)
1	0.72	0.67	0.63	0.33	186.64
2
3
4
5
6