Human Consistency Evaluation on Q-Reasoning (test)
51.4ROUGE-1 ScoreProposed Human-Like Reasoning Framework (detailed)
Evaluation Results
| Method | Links | |
|---|---|---|
| Proposed Human-Like Reasoning Framework (detailed)prompt_template=(c) Detailed Variant Template2025.12 | 51.4 | |
| Proposed Human-Like Reasoning Framework (base)prompt_template=(b) Base Template2025.12 | 51.2 | |
| Q-Insighttraining=RL-based, variant=full2025.12 | 48.7 | |
| Q-Insight-Scoreprompt_template=(a) Baseline Prompt2025.12 | 44.3 | |
| DepictQAtraining=SFT-based2025.12 | 31.8 | |
| Q-Instructtraining=SFT-based2025.12 | 27.9 |