DE-NER : Zero-shot Named Entity Recognition via Dialogue Elicitation of Large Language Models

2026-08-01

AI Summary

This work addresses the limitations of existing large language model–based zero-shot named entity recognition (NER) approaches, which heavily rely on manually crafted prompts and examples, thereby constraining the effective utilization of the model’s inherent knowledge. To overcome this, the study introduces a conversational prompting framework—leveraging dialogue-guided interaction for the first time in zero-shot NER—to naturally elicit the model’s intrinsic entity recognition capabilities without complex prompt engineering. By reformulating the sequence labeling task into a question-answering paradigm, the proposed method substantially reduces dependence on handcrafted exemplars. Experimental results demonstrate that this approach achieves an average F1 score improvement of 3.75% across multiple benchmark datasets, significantly outperforming current zero-shot baselines and confirming its effectiveness and novelty.

Abstract

Recent advancements of zero-shot Named Entity Recognition (NER) establish strong baselines by formulating sequence labeling into question answering where Large Language Models (LLMs) can be naturally adopted. However, existing LLM-based zero-shot NER methods suffer from the limitations of prompt and demonstration engineering. To address these issues with minimal human interventions, we introduce DE-NER, a dialogue elicitation framework which elicits the chatting ability of LLMs to fully extract the knowledge encoded in LLMs. Our experiments demonstrate that the proposed method outperform the competitive baselines in zero-shot settings across multiple benchmarks, with an average improvement of 3.75\% F1 points. Codes are released in https://github.com/kkkenshi/DE-NER.

Problem

• zero-shot Named Entity Recognition
• Large Language Models
• prompt engineering
• demonstration engineering

Innovation

• Zero-shot NER
• Dialogue Elicitation
• Large Language Models
• Prompt Engineering
• Named Entity Recognition