# VicunaNER: Zero/Few-shot Named Entity Recognition using Vicuna

**Authors:** Bin Ji  
**arXiv:** [2305.03253](https://arxiv.org/abs/2305.03253)  
**Date:** 2023-05-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.03253v1/

## Key points


## Summary

In the field of Named Entity Recognition (NER), Large Language Models (LLMs) like ChatGPT have demonstrated impressive capabilities in zero- and few-shot scenarios. To address data leakage and non-reproducible issues associated with online API-based LLMs, the authors propose a new framework called VicunaNER which is based on an open-source LLM called Vicuna. VicunaNER is a two-phase framework that utilizes multi-turn dialogues with Vicuna to recognize entities from texts. The first phase, known as Recognition, leverages Vicuna to identify entities in the text while the second phase, called Re-Recognition, focuses on recognizing entities that were not identified in the first phase. Dialogues for entity checking are incorporated into each phase to filter out incorrect entities. To evaluate the performance of VicunaNER, experiments are conducted on 10 datasets spanning 5 domains for zero/few shot capacity assessment. The experimental results demonstrate superior performance in both settings while comprehensive investigations provide valuable insights into the capabilities and potential applications of Vicuna as an open source LLM for NER tasks.