DE-NER: Improving Zero-Shot NER via Dialogue Elicitation
August 4, 2026
DE-NER uses a dialogue elicitation framework to extract knowledge from LLMs for Named Entity Recognition without manual prompt engineering. The method achieves an average improvement of 3.75% F1 points over existing zero-shot baselines.
HOW THIS AFFECTS YOU
●
builderYou can implement more robust NER pipelines with minimal manual prompt tuning.
●
researcherYou can explore how dialogue-based elicitation outperforms standard sequence labeling tasks.