LELA: an LLM-based Entity Linking Approach with Zero-Shot Domain Adaptation

TL;DR

LELA is a modular zero-shot entity linking method using large language models, achieving high performance across domains without fine-tuning, thus advancing knowledge extraction and question-answering tasks.

Contribution

The paper introduces LELA, a novel zero-shot, domain-adaptive entity linking approach leveraging LLMs without fine-tuning, outperforming existing non-fine-tuned methods.

Findings

LELA performs competitively with fine-tuned approaches.

LELA significantly outperforms other non-fine-tuned methods.

The approach is adaptable to various domains and knowledge bases.

Abstract

Entity linking (mapping ambiguous mentions in text to entities in a knowledge base) is a foundational step in tasks such as knowledge graph construction, question-answering, and information extraction. Our method, LELA, is a modular coarse-to-fine approach that leverages the capabilities of large language models (LLMs), and works with different target domains, knowledge bases and LLMs, without any fine-tuning phase. Our experiments across various entity linking settings show that LELA is highly competitive with fine-tuned approaches, and substantially outperforms the non-fine-tuned ones.