Personalized Response Generation on Review Trained Users
92.3WinrateFSPO + RAT
Evaluation results
| Method | Links | |
|---|---|---|
| FSPO + RATNumber of shots=8, Optimization strategy=Few-Shot Preference Optimization + User Description Rationalization2025.02 | 92.3 | |
| FSPONumber of shots=8, Optimization strategy=Few-Shot Preference Optimization2025.02 | 80.4 | |
| FSPONumber of shots=4, Optimization strategy=Few-Shot Preference Optimization2025.02 | 78.4 | |
| PromptedNumber of shots=8, Optimization strategy=Prompted2025.02 | 69.1 | |
| PromptedNumber of shots=4, Optimization strategy=Prompted2025.02 | 66.6 | |
| Pref-FTNumber of shots=4, Optimization strategy=Preference Fine-Tuning2025.02 | 66.5 | |
| Pref-FTNumber of shots=8, Optimization strategy=Preference Fine-Tuning2025.02 | 65.6 | |
| Llama 3.2 3B InstructNumber of shots=0, Optimization strategy=Instruction-tuned2025.02 | 50 |