Related Experiment Video
Updated: Jul 26, 2026
04:16
Fine-Tuning Large Language Models Using Entity Hallucination Index for Text Summarization
Published on: January 9, 2026
A semantically enhanced two-stage framework for few-shot named entity recognition
Jingguo Ren1, Zhuangzhuang Li2, Yi Yang2
1State Grid Shandong Electric Power Company, Jinan, 250001, Shandong, China.
Scientific Reports
|July 1, 2026
Summary
This study introduces a new framework to improve few-shot named entity recognition (NER) in specialized domains. The semantically enhanced approach boosts performance in low-resource scenarios.
Area of Science:
- Natural Language Processing
- Machine Learning
Background:
- Named Entity Recognition (NER) performance degrades in specialized domains due to scarce annotations and evolving entity types.
- Existing few-shot NER methods struggle with domain shift and unstable type representations from small support sets.
Purpose of the Study:
- To propose a semantically enhanced two-stage framework to improve few-shot NER performance.
- To address challenges of domain shift and unstable type representations in low-resource NER.
Main Methods:
- A boundary-aware span detector is trained using a contrastive objective on a source domain.
- Label-guided hybrid prototypes are constructed by fusing label-text semantics with support-set mentions for target-domain episodes.
Main Results:
- The proposed framework consistently outperforms strong two-stage baselines on cross-domain benchmarks.
- Achieved approximately 1-3 F1 gains in challenging low-resource scenarios on Few-NERD and a power equipment dataset.
Conclusions:
- The semantically enhanced framework effectively improves few-shot NER in specialized, low-resource domains.
- The method offers a robust solution for information extraction challenges in data-scarce environments.