Sentence-level value detection on ValueEval English '24 (test)
33.2Macro F1DeBERTa Ensemble (Baseline + LIWC-22 + Prev-2)
Evaluation results
| Method | Links | |
|---|---|---|
| DeBERTa Ensemble (Baseline + LIWC-22 + Prev-2)Family=Transformer (ensemble), Architecture=Direct ensemble, Voting=soft vote, Threshold (t*)=0.292026.01 | 33.2 | |
| DeBERTa + LIWC-22Family=Transformer (single), Training=tuned, Architecture=Direct2026.01 | 32 | |
| Gemma LLM EnsembleFamily=LLM (ensemble), Architecture=LLM ensemble, Components=Gemma few-shot hier + Gemma QLoRA, Voting=hard vote2026.01 | 27.7 | |
| Gemma-2-9B few-shotFamily=LLM (single), Auxiliary=SBERT presence gate, Architecture=Hier. (presence gate)2026.01 | 23.8 |