Sentiment Transfer on Yelp (test) using STAR metrics
0.519STAR∆GPT-4o-mini
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4o-miniSentiment Level=Very Negative2026.01 | 0.519 | 1.519 | 45.39 | 0.35 | |
| GPT-4o-miniSentiment Level=Very Positive2026.01 | 0.3968 | 4.6032 | 48.98 | 0.41 | |
| SFT + PPO (Rlex + Rcons)Sentiment Level=Neutral, Reward Functions=Rlex + Rcons2026.01 | 0.3904 | 3.3904 | 49.14 | 0.38 | |
| SFTSentiment Level=Neutral2026.01 | 0.3854 | 3.3854 | 46.24 | 0.37 | |
| SFT + PPO (Rsent + Rcons)Sentiment Level=Neutral, Reward Functions=Rsent + Rcons2026.01 | 0.376 | 3.376 | 48.95 | 0.38 | |
| SFT + PPO (Rsent + Rlex + Rcons)Sentiment Level=Neutral, Reward Functions=Rsent + Rlex + Rcons2026.01 | 0.3748 | 3.3748 | 48.89 | 0.39 | |
| GPT-4o-miniSentiment Level=Positive2026.01 | 0.3652 | 4.3652 | 49.78 | 0.3 | |
| SFTSentiment Level=Negative2026.01 | 0.3338 | 2.3338 | 45.88 | 0.38 | |
| SFT + PPO (Rsent + Rcons)Sentiment Level=Negative, Reward Functions=Rsent + Rcons2026.01 | 0.3254 | 2.3254 | 48.55 | 0.38 | |
| SFT + PPO (Rsent + Rlex + Rcons)Sentiment Level=Negative, Reward Functions=Rsent + Rlex + Rcons2026.01 | 0.3234 | 2.3234 | 48.49 | 0.4 | |
| SFT + PPO (Rlex + Rcons)Sentiment Level=Negative, Reward Functions=Rlex + Rcons2026.01 | 0.3146 | 2.3146 | 48.8 | 0.38 | |
| GPT-4o-miniSentiment Level=Average2026.01 | 0.3063 | — | 47.57 | 0.33 | |
| SFTSentiment Level=Average2026.01 | 0.2234 | — | 45.18 | 0.43 | |
| SFT + PPO (Rlex + Rcons)Sentiment Level=Average, Reward Functions=Rlex + Rcons2026.01 | 0.2203 | — | 47.97 | 0.44 | |
| SFT + PPO (Rsent + Rlex + Rcons)Sentiment Level=Average, Reward Functions=Rsent + Rlex + Rcons2026.01 | 0.2186 | — | 47.8 | 0.45 | |
| SFT + PPO (Rsent + Rcons)Sentiment Level=Average, Reward Functions=Rsent + Rcons2026.01 | 0.2185 | — | 48.15 | 0.44 | |
| SFT + PPO (Rsent + Rlex + Rcons)Sentiment Level=Very Negative, Reward Functions=Rsent + Rlex + Rcons2026.01 | 0.2114 | 1.2114 | 46.18 | 0.51 | |
| SFTSentiment Level=Very Negative2026.01 | 0.2012 | 1.2012 | 43.68 | 0.5 | |
| SFT + PPO (Rlex + Rcons)Sentiment Level=Very Negative, Reward Functions=Rlex + Rcons2026.01 | 0.2004 | 1.2004 | 46.19 | 0.5 | |
| GPT-4o-miniSentiment Level=Negative2026.01 | 0.1982 | 1.8018 | 47.11 | 0.38 | |
| SFT + PPO (Rsent + Rcons)Sentiment Level=Very Negative, Reward Functions=Rsent + Rcons2026.01 | 0.1954 | 1.1954 | 46.16 | 0.5 | |
| SFT + PPO (Rsent + Rlex + Rcons)Sentiment Level=Positive, Reward Functions=Rsent + Rlex + Rcons2026.01 | 0.1462 | 4.1462 | 49.04 | 0.33 | |
| SFTSentiment Level=Positive2026.01 | 0.1382 | 4.1382 | 46.2 | 0.32 | |
| SFT + PPO (Rlex + Rcons)Sentiment Level=Positive, Reward Functions=Rlex + Rcons2026.01 | 0.137 | 4.137 | 49.21 | 0.32 | |
| SFT + PPO (Rsent + Rcons)Sentiment Level=Positive, Reward Functions=Rsent + Rcons2026.01 | 0.133 | 4.133 | 49.21 | 0.32 | |
| SFT + PPO (Rsent + Rcons)Sentiment Level=Very Positive, Reward Functions=Rsent + Rcons2026.01 | 0.0626 | 4.9374 | 47.88 | 0.6 | |
| SFT + PPO (Rlex + Rcons)Sentiment Level=Very Positive, Reward Functions=Rlex + Rcons2026.01 | 0.059 | 4.941 | 46.53 | 0.6 | |
| SFTSentiment Level=Very Positive2026.01 | 0.0584 | 4.9416 | 43.92 | 0.6 | |
| GPT-4o-miniSentiment Level=Neutral2026.01 | 0.0524 | 2.9476 | 46.58 | 0.23 | |
| SFT + PPO (Rsent + Rlex + Rcons)Sentiment Level=Very Positive, Reward Functions=Rsent + Rlex + Rcons2026.01 | 0.0374 | 4.9374 | 46.39 | 0.62 |