Follow-up question generation on FollowupBench
82RecallKG-Followup + Active ICL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| KG-Followup + Active ICLBackbone=Claude Sonnet, Evaluation protocol=Active ICL2026.03 | 82 | 40 | |
| KG-FollowupBackbone=Claude Sonnet, Evaluation protocol=KG-Followup2026.03 | 81 | 40 | |
| KG-Followup + Random ICLBackbone=Claude Sonnet, Evaluation protocol=Random ICL2026.03 | 81 | 40 | |
| KG-Followup + Active ICLBackbone=Claude Haiku, Evaluation protocol=Active ICL2026.03 | 78 | 40 | |
| KG-FollowupBackbone=Claude Haiku, Evaluation protocol=KG-Followup2026.03 | 77 | 40 | |
| KG-Followup + Random ICLBackbone=Claude Haiku, Evaluation protocol=Random ICL2026.03 | 77 | 40 | |
| MedGemma-27bBackbone=MedGemma-27b2026.03 | 74 | 40 | |
| FollowupQBackbone=Claude Sonnet, Evaluation protocol=FollowupQ2026.03 | 74 | 40 | |
| FollowupQBackbone=Claude Haiku, Evaluation protocol=FollowupQ2026.03 | 73 | 40 | |
| Zero-Shot-kBackbone=Claude Haiku, Evaluation protocol=Zero-Shot-k2026.03 | 72 | 40 | |
| Zero-Shot-kBackbone=Claude Sonnet, Evaluation protocol=Zero-Shot-k2026.03 | 72 | 40 | |
| GPT-4oBackbone=GPT-4o2026.03 | 67 | 40 | |
| Zero-Shot-UBackbone=Claude Haiku, Evaluation protocol=Zero-Shot-U2026.03 | 54 | 10 | |
| Zero-Shot-UBackbone=Claude Sonnet, Evaluation protocol=Zero-Shot-U2026.03 | 53 | 10 |