Contextual Augmentation for Entity Linking using Large Language Models

TL;DR

The paper tackles Entity Linking by unifying NER and ED in a single, fine-tuned framework and boosting disambiguation with LLM-based context augmentation. It deploys a T5-based architecture paired with a LlaMA3-driven expansion of mentions to richer Wikipedia-like titles, alongside mitigation for hallucinations via a title-URI dictionary. Extensive experiments across multiple out-of-domain benchmarks show state-of-the-art performance in many cases, with detailed ablations demonstrating when augmentation helps most. The work reduces resource demands relative to separate NER and ED models and provides practical guidance on augmentation strategies, ultimately contributing to more robust cross-domain EL systems while acknowledging limitations and avenues for future benchmarks.

Abstract

Entity Linking involves detecting and linking entity mentions in natural language texts to a knowledge graph. Traditional methods use a two-step process with separate models for entity recognition and disambiguation, which can be computationally intensive and less effective. We propose a fine-tuned model that jointly integrates entity recognition and disambiguation in a unified framework. Furthermore, our approach leverages large language models to enrich the context of entity mentions, yielding better performance in entity disambiguation. We evaluated our approach on benchmark datasets and compared with several baselines. The evaluation results show that our approach achieves state-of-the-art performance on out-of-domain datasets.

Figures (1)

• Figure 1: The architecture of our approach, including Jointly Fine-tuning and LLM-based Augmentation