Real Play Prediction on Large-scale Video Search Dataset (offline experiments)
0.85AUCTRM-RankMixer
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TRM-RankMixerDense Param=352M, Sparse Param=5.07T, FLOPs (bs=4096)=14.66T2026.01 | 0.85 | 0.7 | |
| DLRM-MLPMethod Category=ID Baseline, Dense Param=7M, Sparse Param=7.52T, FLOPs (bs=4096)=0.13T2026.01 | 0.8413 | 0.8604 | |
| TRM-Pure TransformerDense Param=341M, Sparse Param=5.07T, FLOPs (bs=4096)=12.17T2026.01 | 0.81 | 0.68 | |
| RankMixerMethod Category=ID-based, Dense Param=335M, Sparse Param=7.52T, FLOPs (bs=4096)=13.20T2026.01 | 0.76 | 0.63 | |
| SEMIDMethod Category=Token-based, Dense Param=345M, Sparse Param=5.08T, FLOPs (bs=4096)=14.23T2026.01 | 0.75 | 0.61 | |
| Pure TransformerMethod Category=ID-based, Dense Param=326M, Sparse Param=7.52T, FLOPs (bs=4096)=11.05T2026.01 | 0.69 | 0.56 | |
| OneRec-tokenMethod Category=Token-based, Dense Param=342M, Sparse Param=5.06T, FLOPs (bs=4096)=13.98T2026.01 | 0.64 | 0.54 | |
| WuKongMethod Category=ID-based, Dense Param=355M, Sparse Param=7.52T, FLOPs (bs=4096)=18.96T2026.01 | 0.59 | 0.45 | |
| Tiger-tokenMethod Category=Token-based, Dense Param=338M, Sparse Param=5.06T, FLOPs (bs=4096)=13.89T2026.01 | 0.58 | 0.45 | |
| DHENMethod Category=ID-based, Dense Param=242M, Sparse Param=7.52T, FLOPs (bs=4096)=8.42T2026.01 | 0.33 | 0.31 | |
| DCN-v2Method Category=ID-based, Dense Param=78M, Sparse Param=7.52T, FLOPs (bs=4096)=1.92T2026.01 | 0.15 | 0.1 |