Logical Refinement of Natural Language Explanations on WorldTree
9Initial PerformanceFaithful-Refiner
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Faithful-RefinerLLM=GPT-4o2025.05 | 9 | 56 | 3.86 | 26.16 | |
| Faithful-RefinerLLM=Deepseek-V32025.05 | 7 | 73 | 3.55 | 25.3 | |
| Faithful-RefinerLLM=Llama3.1-70b2025.05 | 6 | 52 | 4.62 | 35.72 | |
| Faithful-RefinerLLM=GPT-4o-mini2025.05 | 5 | 47 | 4.75 | 36.5 | |
| Explanation-RefinerLLM=Deepseek-V32025.05 | 3 | 31 | 4.52 | 42.64 | |
| Explanation-RefinerLLM=Llama3.1-70b2025.05 | 2 | 15 | 5.23 | 51.61 | |
| Explanation-RefinerLLM=GPT-4o2025.05 | 2 | 13 | 4.18 | 39.26 | |
| Explanation-RefinerLLM=GPT-4o-mini2025.05 | 0 | 4 | 5 | 46.12 |