Dialogue Analysis on DA
92.5R MetricHalluClean
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HalluCleanProtocol=Detection → Revision (with rationale), Backbone=DeepSeek-V32025.11 | 92.5 | 86 | |
| HalluCleanProtocol=Detection → Revision (with rationale), Backbone=GPT-3.5-turbo2025.11 | 89 | 83 | |
| DeepSeek-V3Protocol=Direct Ask, Backbone=DeepSeek-V32025.11 | 86.5 | 78 | |
| GPT-4o-miniProtocol=Direct Ask, Backbone=GPT-4o-mini2025.11 | 84.5 | 79.5 | |
| ChatProtectProtocol=Detection → Revision (with rationale), Backbone=GPT-3.5-turbo2025.11 | 79.5 | 74 | |
| Llama-3-70BProtocol=Direct Ask, Backbone=Llama-3-70B2025.11 | 74.5 | 68.5 | |
| DeepSeek-R1Protocol=Direct Ask, Backbone=DeepSeek-R12025.11 | 74 | 67.5 | |
| GPT-3.5-turboProtocol=Direct Ask, Backbone=GPT-3.5-turbo2025.11 | 57.5 | 53 | |
| Step-by-StepProtocol=Detection → Revision (with rationale), Backbone=GPT-3.5-turbo2025.11 | 54.5 | 51.5 | |
| Plan-and-SolveProtocol=Detection → Revision (with rationale), Backbone=GPT-3.5-turbo2025.11 | 11.5 | 10.5 |