POI Attribute Acquisition (Task-oriented Dialogue) on DuIVRS logs Dgeneral 1.0 (test)
73.83Completion RateHybridLLMs
Evaluation results
| Method | Links | |
|---|---|---|
| HybridLLMsLLM-S=Qwen2.5-1.5B, LLM-L=Qwen2.5-7B, Black-box LLM=GPT-4o2026.05 | 73.83 | |
| DuIVRS-2LLM-S=ERNIE-Bot-tiny, LLM-L=ERNIE-Bot-turbo, Black-box LLM=ERNIE 4.02026.05 | 73.7 | |
| LLM-DM2026.05 | 67.4 | |
| GPT-4o2026.05 | 64.81 | |
| DeepSeek-V32026.05 | 63.87 | |
| DuIVRS-12026.05 | 62.99 | |
| w/o-DA2026.05 | 62.36 | |
| Direct-SFT2026.05 | 60.33 | |
| w/o-CoT2026.05 | 44.41 |