CTR Prediction on Industrial Dataset (Douyin Search) (test)
0.6489AUCHyFormer
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HyFormerSequence Modeling=HyFormer, Feature Interaction=Unified-Block, Params (×10^6)=418, FLOPs (×10^12)=3.92026.01 | 0.6489 | 0.17 | |
| MTGR/OneTrans (w/ Full Transformer)Sequence Modeling=MTGR/OneTrans (w/ Full Transformer), Feature Interaction=Unified-Block, Params (×10^6)=450, FLOPs (×10^12)=21.92026.01 | 0.6483 | 0.08 | |
| Full Transformer + RankMixerSequence Modeling=Full Transformer, Feature Interaction=RankMixer, Params (×10^6)=388, FLOPs (×10^12)=6.62026.01 | 0.6481 | 0.05 | |
| MTGR/OneTrans (w/ LONGER)Sequence Modeling=MTGR/OneTrans (w/ LONGER), Feature Interaction=Unified-Block, Params (×10^6)=406, FLOPs (×10^12)=6.62026.01 | 0.648 | 0.03 | |
| LONGER + RankMixerSequence Modeling=LONGER, Feature Interaction=RankMixer, Params (×10^6)=386, FLOPs (×10^12)=3.52026.01 | 0.6478 | — | |
| Full Transformer + Full TransformerSequence Modeling=Full Transformer, Feature Interaction=Full Transformer, Params (×10^6)=418, FLOPs (×10^12)=9.32026.01 | 0.6474 | -0.06 | |
| LONGER + Full TransformerSequence Modeling=LONGER, Feature Interaction=Full Transformer, Params (×10^6)=416, FLOPs (×10^12)=6.22026.01 | 0.6472 | -0.09 | |
| Full Transformer + WukongSequence Modeling=Full Transformer, Feature Interaction=Wukong, Params (×10^6)=387, FLOPs (×10^12)=8.32026.01 | 0.6468 | -0.15 | |
| LONGER + WukongSequence Modeling=LONGER, Feature Interaction=Wukong, Params (×10^6)=385, FLOPs (×10^12)=5.22026.01 | 0.6465 | -0.2 |