Human Evaluation of Explanations on DRUID
1.917Helpfulness (MAR)CLUE-Span
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CLUE-SpanBase Model=Qwen2.5-14B-Instruct, Prompting Strategy=CLUE-Span2025.05 | 1.917 | 1.75 | 1.983 | 1.75 | 1.9 | |
| PromptBaselineBase Model=Qwen2.5-14B-Instruct, Prompting Strategy=PromptBaseline2025.05 | 1.9 | 1.717 | 1.983 | 1.767 | 1.9 | |
| CLUE-Span+SteeringBase Model=Qwen2.5-14B-Instruct, Prompting Strategy=CLUE-Span+Steering2025.05 | 1.767 | 1.617 | 1.683 | 1.617 | 1.817 | |
| CLUEBackbone=Qwen2.5-14B-Instruct2025.05 | 0.688 | 0.691 | 0.739 | 0.717 | 0.688 | |
| PromptBaselineBackbone=Qwen2.5-14B-Instruct2025.05 | 0.312 | 0.309 | 0.261 | 0.283 | 0.313 |