Feature Dual Supervision Model for the Searches of Online Advertising Audiences

Table 3

Results of comparison experiment on Lookalike and MovieLens.

Datasets	Lookalike				MovieLens
Model	GAUC	Logloss	Precision@5%	Recall@5%	GAUC	Logloss	Precision@50%	Recall@50%
FM	0.7076	0.1661	0.2225	0.2295	0.6974	0.6171	0.7188	0.6214
MLP	0.7291	0.1616	0.2403	0.2479	0.7003	0.6119	0.7193	0.6211
DeepFM	0.7272	0.1620	0.2385	0.2462	0.7001	0.6094	0.7192	0.6210
PNN	0.7165	0.1642	0.2280	0.2351	0.7016	0.6112	0.7205	0.6218
Wide&Deep	0.7285	0.1617	0.2399	0.2478	0.6975	0.6110	0.7171	0.6183
Two-tower	0.7221	0.1725	0.2393	0.2470	0.7011	0.6091	0.7191	0.6208
DCN	0.7287	0.1617	0.2401	0.2477	0.6982	0.6162	0.7184	0.6199
AFN+	0.7170	0.1648	0.2214	0.2282	0.7002	0.6097	0.7209	0.6228
FDSM	0.7366	0.1601	0.2471	0.2550	0.7072	0.6072	0.7254	0.6270

The best performances are in bold.