Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OneNet: A Fine-Tuning Free Framework for Few-Shot Entity Linking via Large Language Model Prompting

Xukai Liu, Ye Liu, Kai Zhang, Kehang Wang, Qi Liu, Enhong Chen

TL;DR

Entity Linking in few-shot settings is challenged by data scarcity, long input spans, and hallucinations. OneNet tackles this by prompting three LLM-based modules—ERP for input reduction, DEL for balanced contextual and prior reasoning, and ECJ for consistency-based disagreement resolution—without fine-tuning. Empirical results across seven benchmarks show OneNet achieving state-of-the-art performance, with ablations underscoring the value of dual perspectives and the consistency judger. This approach has practical impact by enabling robust EL in domain-specific or low-resource scenarios using off-the-shelf LLMs with carefully crafted prompts.

Abstract

Entity Linking (EL) is the process of associating ambiguous textual mentions to specific entities in a knowledge base. Traditional EL methods heavily rely on large datasets to enhance their performance, a dependency that becomes problematic in the context of few-shot entity linking, where only a limited number of examples are available for training. To address this challenge, we present OneNet, an innovative framework that utilizes the few-shot learning capabilities of Large Language Models (LLMs) without the need for fine-tuning. To the best of our knowledge, this marks a pioneering approach to applying LLMs to few-shot entity linking tasks. OneNet is structured around three key components prompted by LLMs: (1) an entity reduction processor that simplifies inputs by summarizing and filtering out irrelevant entities, (2) a dual-perspective entity linker that combines contextual cues and prior knowledge for precise entity linking, and (3) an entity consensus judger that employs a unique consistency algorithm to alleviate the hallucination in the entity linking reasoning. Comprehensive evaluations across seven benchmark datasets reveal that OneNet outperforms current state-of-the-art entity linking methods.

OneNet: A Fine-Tuning Free Framework for Few-Shot Entity Linking via Large Language Model Prompting

TL;DR

Entity Linking in few-shot settings is challenged by data scarcity, long input spans, and hallucinations. OneNet tackles this by prompting three LLM-based modules—ERP for input reduction, DEL for balanced contextual and prior reasoning, and ECJ for consistency-based disagreement resolution—without fine-tuning. Empirical results across seven benchmarks show OneNet achieving state-of-the-art performance, with ablations underscoring the value of dual perspectives and the consistency judger. This approach has practical impact by enabling robust EL in domain-specific or low-resource scenarios using off-the-shelf LLMs with carefully crafted prompts.

Abstract

Entity Linking (EL) is the process of associating ambiguous textual mentions to specific entities in a knowledge base. Traditional EL methods heavily rely on large datasets to enhance their performance, a dependency that becomes problematic in the context of few-shot entity linking, where only a limited number of examples are available for training. To address this challenge, we present OneNet, an innovative framework that utilizes the few-shot learning capabilities of Large Language Models (LLMs) without the need for fine-tuning. To the best of our knowledge, this marks a pioneering approach to applying LLMs to few-shot entity linking tasks. OneNet is structured around three key components prompted by LLMs: (1) an entity reduction processor that simplifies inputs by summarizing and filtering out irrelevant entities, (2) a dual-perspective entity linker that combines contextual cues and prior knowledge for precise entity linking, and (3) an entity consensus judger that employs a unique consistency algorithm to alleviate the hallucination in the entity linking reasoning. Comprehensive evaluations across seven benchmark datasets reveal that OneNet outperforms current state-of-the-art entity linking methods.

Paper Structure

This paper contains 35 sections, 7 equations, 8 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Entity Linking
  4. Large Language Models
  5. Preliminary
  6. Few-shot Entity Linking
  7. Entity Linking with LLMs
  8. Method
  9. An Overview of One-Net
  10. Entity Reduction Processor
  11. Entity Representation Summarizer
  12. Candidate Entity Filter
  13. Dual-perspective Entity Linker
  14. Contextual Entity Linker
  15. Prior Entity Linker
  16. ...and 20 more sections

Figures (8)

  • Figure 1: An example of entity linking across various scenarios, where mention is bolded in red.
  • Figure 2: The illustration of OneNet framework, which contains three distinct modules: (a) Entity Reduction Processor (ERP), (b) Dual-perspective Entity Linker (DEL), and (c) Entity Consensus Judger (ECJ).
  • Figure 3: The Illustration of General Prompt Structure
  • Figure 4: Comparison of Prior, Context, and Merge
  • Figure 5: Case study of OneNet. Key information in context makes the contextual linker reason correctly.
  • ...and 3 more figures