Therapeutic Dialogue Generation on MESC (test)
0.0111BLEUReinforcement Learning - SFT
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Reinforcement Learning - SFTTraining=Reinforcement learning fine-tuned after SFT2025.11 | 0.0111 | 0.1397 | 0.0213 | 0.1317 | 0.0581 | |
| SFT (Therapist emotions)Emotion Context=Inclusive, Training=Supervised Fine-Tuning2025.11 | 0.0108 | 0.1164 | 0.0175 | 0.1076 | 0.0485 | |
| SFT (No therapist emotions)Emotion Context=None, Training=Supervised Fine-Tuning2025.11 | 0.0106 | 0.116 | 0.0159 | 0.1058 | 0.0433 | |
| GPT-2Description=Baseline model2025.11 | 0.0021 | 0.0478 | 0.0064 | 0.0408 | 0.0691 |