Introduction
Related work
User interest
Learning feature interactions
Self-attention and residual networks
IARM model
Overview
Input layer
Embedding layer
Interest acquisition layer
Interaction layer
Output layer
Training
Experiment
Experimental setup
Experimental data set
Data | Samples | Fields | Features (sparse) |
---|---|---|---|
Criteo | 45,840,617 | 39 | 998,960 |
Avazu | 40,428,967 | 23 | 1,544,488 |
Movielens-1M | 739,012 | 7 | 3529 |
Model | Criteo | Movielens-1M | Avazu | |||
---|---|---|---|---|---|---|
AUC | LOSS | AUC | LOSS | AUC | LOSS | |
FM | 0.6869 | 0.5286 | 0.5347 | 0.4462 | 0.5437 | 0.6221 |
Weidedeep | 0.7066 | 0.4827 | 0.8328 | 0.3334 | 0.7424 | 0.4117 |
Deepfm | 0.7283 | 0.4707 | 0.8340 | 0.3346 | 0.7461 | 0.4041 |
AFM | 0.7220 | 0.4754 | 0.8295 | 0.3358 | 0.7567 | 0.4012 |
DCN | 0.7094 | 0.4920 | 0.8249 | 0.3393 | 0.7349 | 0.4139 |
NFM | 0.7027 | 0.5645 | 0.8297 | 0.3357 | 0.7400 | 0.4179 |
PNN | 0.7084 | 0.4870 | 0.8312 | 0.3353 | 0.7374 | 0.4096 |
Autoint | 0.7060 | 0.6049 | 0.8362 | 0.3393 | 0.7450 | 0.4496 |
Deepcrosing | 0.7375 | 0.4732 | 0.8373 | 0.3305 | 0.7572 | 0.3989 |
IARM | 0.7545 | 0.4830 | 0.8386 | 0.3296 | 0.7652 | 0.3994 |
Data sets | Model | AUC | LOSS |
---|---|---|---|
Criteo | IARM | 0.7545 | 0.4830 |
IARM* | 0.7371 | 0.4965 | |
Avazu | IARM | 0.7652 | 0.3994 |
IARM* | 0.7591 | 0.3995 | |
Movielens | IARM | 0.8386 | 0.3296 |
IARM* | 0.8349 | 0.3344 |
Data sets | Model | AUC | LOSS |
---|---|---|---|
Criteo | IARM | 0.7545 | 0.4830 |
IRM | 0.7497 | 0.4902 | |
Avazu | IARM | 0.7652 | 0.3994 |
IRM | 0.7637 | 0.4019 | |
Movielens | IARM | 0.8386 | 0.3296 |
IRM | 0.8345 | 0.3292 |
Data sets | Model | AUC | LOSS |
---|---|---|---|
Criteo | IARM | 0.7545 | 0.4830 |
IARM- | 0.7511 | 0.4822 | |
Avazu | IARM | 0.7652 | 0.3994 |
IARM- | 0.7621 | 0.4086 | |
Movielens | IARM | 0.8386 | 0.3296 |
IARM- | 0.8236 | 0.3422 |
Model | Criteo | |||||
---|---|---|---|---|---|---|
Proportion | 0.2 | 0.3 | AVG. changes | |||
AUC | LOSS | AUC | LOSS | AUC | LOSS | |
FM | 0.6869 | 0.5286 | 0.6785 | 0.5368 | − 0.0084 | + 0.0082 |
Weidedeep | 0.7066 | 0.4827 | 0.7007 | 0.4846 | − 0.0059 | + 0.0019 |
Deepfm | 0.7283 | 0.4707 | 0.7249 | 0.4792 | − 0.0034 | + 0.0085 |
AFM | 0.7220 | 0.4754 | 0.7084 | 0.4877 | − 0.0136 | + 0.0123 |
DCN | 0.7094 | 0.4920 | 0.7042 | 0.5010 | − 0.0052 | + 0.009 |
NFM | 0.7027 | 0.5645 | 0.7005 | 0.5654 | − 0.0022 | + 0.0009 |
PNN | 0.7084 | 0.4870 | 0.7013 | 0.4979 | − 0.0071 | + 0.0109 |
Autoint | 0.7060 | 0.6049 | 0.6921 | 0.6552 | − 0.0139 | + 0.0503 |
Deepcrosing | 0.7375 | 0.4732 | 0.7356 | 0.4792 | −0.0019 | + 0.006 |
IARM | 0.7545 | 0.4830 | 0.7527 | 0.4993 | − 0.0018 | + 0.0163 |