Actionable Suggestion Extraction on Manual evaluation set 1.0 (test)
92BERTScoreHybrid pipeline
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Hybrid pipelineModel Type=Supervised-LLM Hybrid2026.01 | 92 | 89 | 32 | 68 | |
| Prompt-only LLMBackbone=Gemma-32026.01 | 87 | 84 | 56 | 70 | |
| T5-base (span)Backbone=T5-base, Mode=Span-based2026.01 | 78 | 76 | 72 | 73 | |
| Rule-basedModel Type=Rule-based extractor2026.01 | 46 | 44 | 42 | 45 |