ARTFEED — Contemporary Art Intelligence

LELA: A Zero-Shot LLM-Based Entity Linking Framework

ai-technology · 2026-05-27

A practical Python library has been developed by researchers to enhance LELA, a modular and domain-agnostic method for entity disambiguation based on LLM. This library incorporates zero-shot Named Entity Recognition (NER), offering a comprehensive end-to-end solution for entity linking in real-world applications. Experimental findings confirm the effectiveness and reliability of LELA in various entity linking scenarios. Additionally, a demonstration is available, enabling users to evaluate the system using their own text inputs.

Key facts

  • LELA is an end-to-end LLM-based entity linking framework.
  • It supports zero-shot domain adaptation.
  • The framework integrates zero-shot Named Entity Recognition (NER).
  • It is provided as a practical Python library.
  • Experimental results show performance and robustness.
  • A demo is available for user input testing.
  • The paper is categorized under Computer Science > Artificial Intelligence.
  • The submission is on arXiv.

Entities

Institutions

  • arXiv

Sources