DIN: It seems that the feature engineering(negative sampling) of paper used for amazon book seems bad. I try hard but the auc of test cannot reach the 0.811 on amazon book.
Multi tasks
Model
dataset
loss_func
metrics
state
MMOE
census-income
BCEWithLogitsLoss
income-AUC: 0.9061marry-AUC: 0.9637
π’
ESMM
census-income
BCEWithLogitsLoss
income-ctr-AUC: 0.9242ctcvr-AUC: 0.9122
π’
About
π΄ PyTorch Implementation of classic Recommender System Models.