Advertiser prediction on V 1 (offline evaluation)
24.8Recall@1SID-enabled SFT (1-advertiser prompt) + GRPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SID-enabled SFT (1-advertiser prompt) + GRPOTraining protocol=SFT + GRPO, Prompt format=1-advertiser prompt, Feature enhancement=Semantic ID (SID) enabled2026.05 | 24.8 | 51.5 | 75.5 | |
| SFT (1-advertiser prompt) + GRPOTraining protocol=SFT + GRPO, Prompt format=1-advertiser prompt2026.05 | 22.3 | 46.1 | 68.3 | |
| SFT (1-advertiser prompt)Training protocol=SFT, Prompt format=1-advertiser prompt2026.05 | 21.4 | 41.3 | 50.1 | |
| SFT (20-advertiser prompt)Training protocol=SFT, Prompt format=20-advertiser prompt2026.05 | 15.6 | 31.4 | 45.6 | |
| Zero-shot promptingmode=zero-shot2026.05 | 11.7 | 30.1 | 42.2 |