Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

P-ICL: Point In-Context Learning for Named Entity Recognition with Large Language Models

Guochao Jiang, Zepeng Ding, Yuchen Shi, Deqing Yang

TL;DR

A new prompting framework P-ICL is proposed to better achieve NER with LLMs, in which some point entities are leveraged as the auxiliary information to recognize each entity type.

Abstract

In recent years, the rise of large language models (LLMs) has made it possible to directly achieve named entity recognition (NER) without any demonstration samples or only using a few samples through in-context learning (ICL). However, standard ICL only helps LLMs understand task instructions, format and input-label mapping, but neglects the particularity of the NER task itself. In this paper, we propose a new prompting framework P-ICL to better achieve NER with LLMs, in which some point entities are leveraged as the auxiliary information to recognize each entity type. With such significant information, the LLM can achieve entity classification more precisely. To obtain optimal point entities for prompting LLMs, we also proposed a point entity selection method based on K-Means clustering. Our extensive experiments on some representative NER benchmarks verify the effectiveness of our proposed strategies in P-ICL and point entity selection.

P-ICL: Point In-Context Learning for Named Entity Recognition with Large Language Models

TL;DR

A new prompting framework P-ICL is proposed to better achieve NER with LLMs, in which some point entities are leveraged as the auxiliary information to recognize each entity type.

Abstract

In recent years, the rise of large language models (LLMs) has made it possible to directly achieve named entity recognition (NER) without any demonstration samples or only using a few samples through in-context learning (ICL). However, standard ICL only helps LLMs understand task instructions, format and input-label mapping, but neglects the particularity of the NER task itself. In this paper, we propose a new prompting framework P-ICL to better achieve NER with LLMs, in which some point entities are leveraged as the auxiliary information to recognize each entity type. With such significant information, the LLM can achieve entity classification more precisely. To obtain optimal point entities for prompting LLMs, we also proposed a point entity selection method based on K-Means clustering. Our extensive experiments on some representative NER benchmarks verify the effectiveness of our proposed strategies in P-ICL and point entity selection.
Paper Structure (17 sections, 4 equations, 3 figures, 7 tables)

This paper contains 17 sections, 4 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Methodology
  5. Standard ICL
  6. Point Entity
  7. P-ICL: Point ICL
  8. Point Entity Selection
  9. Experiments
  10. Datasets
  11. Overall Performance
  12. Effect of Point Entity Supervision
  13. Effect of Point Entity Number
  14. Effect of Point Entity Selection
  15. Conclusion
  16. ...and 2 more sections

Figures (3)

  • Figure 1: The overview of LLM-based NER using the standard ICL and the P-ICL proposed in this paper. Standard ICL cannot provide LLM with sufficient entity type information and entity classification. Comparatively, P-ICL can make up for this shortcoming by providing some point entities per entity type.
  • Figure 2: The overview of P-ICL with point entity selection. These entities are first given a vector representation using a BERT-like encoder. Then, the K-Means clustering method and nearest neighbor decoding strategy are used to identify the point entities for each entity type. The word cloud figure on the right displays the outcomes of this selection process. For instance, London, China, and California emerge as central entities in their clusters and are identified as point entities for LOC.
  • Figure 3: The impact of varying point entity numbers on model performance. Performance of LLaMA-3-70B on the CoNLL2003 dataset.