Top-10 Recommendation on Amazon Beauty (test)
0.2977Recall@1RL-rejection-KP
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RL-rejection-KPPost-training stage=RL, Reasoning trace generation strategy=rejection, Domain knowledge priming=KP, Backbone=Qwen3-8B2026.02 | 0.2977 | — | — | — | — | 0.746 | 0.9563 | 0.5234 | 0.605 | — | — | |
| RL-targeted-noKPPost-training stage=RL, Reasoning trace generation strategy=targeted, Domain knowledge priming=noKP, Backbone=Qwen3-8B2026.02 | 0.2932 | — | — | — | — | 0.7318 | 0.9534 | 0.5172 | 0.6038 | — | — | |
| RL-rejection-noKPPost-training stage=RL, Reasoning trace generation strategy=rejection, Domain knowledge priming=noKP, Backbone=Qwen3-8B2026.02 | 0.2915 | — | — | — | — | 0.733 | 0.958 | 0.5172 | 0.6036 | — | — | |
| RL-targeted-KPPost-training stage=RL, Reasoning trace generation strategy=targeted, Domain knowledge priming=KP, Backbone=Qwen3-8B2026.02 | 0.2898 | — | — | — | — | 0.7221 | 0.9534 | 0.5101 | 0.5998 | — | — | |
| Pre-rank (MTL)Post-training stage=Pre-rank, Training=Multi-task learning (MTL), Backbone=Qwen3-8B2026.02 | 0.2892 | — | — | — | — | 0.7227 | 0.9534 | 0.5101 | 0.5997 | — | — | |
| RL-zeroshotPost-training stage=RL, Reasoning trace generation strategy=zeroshot, Backbone=Qwen3-8B2026.02 | 0.2801 | — | — | — | — | 0.7074 | 0.9574 | 0.4982 | 0.5926 | — | — | |
| SFT-rejection-KPPost-training stage=SFT, Reasoning trace generation strategy=rejection, Domain knowledge priming=KP, Backbone=Qwen3-8B2026.02 | 0.2784 | — | — | — | — | 0.7091 | 0.9483 | 0.497 | 0.5907 | — | — | |
| SFT-targeted-noKPPost-training stage=SFT, Reasoning trace generation strategy=targeted, Domain knowledge priming=noKP, Backbone=Qwen3-8B2026.02 | 0.2761 | — | — | — | — | 0.7272 | 0.95 | 0.5087 | 0.5964 | — | — | |
| SFT-rejection-noKPPost-training stage=SFT, Reasoning trace generation strategy=rejection, Domain knowledge priming=noKP, Backbone=Qwen3-8B2026.02 | 0.2682 | — | — | — | — | 0.7216 | 0.9574 | 0.5003 | 0.5904 | — | — | |
| SFT-targeted-KPPost-training stage=SFT, Reasoning trace generation strategy=targeted, Domain knowledge priming=KP, Backbone=Qwen3-8B2026.02 | 0.2006 | — | — | — | — | 0.725 | 0.9545 | 0.4759 | 0.5647 | — | — | |
| AlphaFreeUF (user-free)=true, IF (ID-free)=true, GF (GNN-free)=true2026.03 | — | — | — | — | — | — | — | — | — | 0.0361 | 0.02 | |
| FISM-BPRUF (user-free)=true, IF (ID-free)=false, GF (GNN-free)=true2026.03 | — | — | — | — | — | — | — | — | — | 0.0079 | 0.0044 | |
| LightGCNUF (user-free)=false, IF (ID-free)=false, GF (GNN-free)=false2026.03 | — | — | — | — | — | — | — | — | — | 0.0201 | 0.011 | |
| MF-BPRUF (user-free)=false, IF (ID-free)=false, GF (GNN-free)=true2026.03 | — | — | — | — | — | — | — | — | — | 0.0063 | 0.0032 | |
| XSimGCLUF (user-free)=false, IF (ID-free)=false, GF (GNN-free)=false2026.03 | — | — | — | — | — | — | — | — | — | 0.0253 | 0.014 |