GLiDRE: Generalist Lightweight model for Document-level Relation Extraction

TL;DR

GLiDRE is a compact, versatile model for document-level relation extraction that excels in low-resource and few-shot scenarios, outperforming existing methods and setting new benchmarks.

Contribution

We introduce GLiDRE, a novel lightweight model that effectively handles document-level relation extraction in limited data settings, advancing the state-of-the-art.

Findings

Outperforms existing methods in low-resource scenarios

Establishes new state-of-the-art in few-shot relation extraction

Effective in both supervised and meta-learning settings

Abstract

Relation Extraction (RE) is a fundamental task in Natural Language Processing, and its document-level variant poses significant challenges, due to complex interactions between entities across sentences. While supervised models have achieved strong results in fully resourced settings, their behavior with limited training data remains insufficiently studied. We introduce GLiDRE, a new compact model for document-level relation extraction, designed to work efficiently in both supervised and few-shot settings. Experiments in both low-resource supervised training and few-shot meta-learning benchmarks show that our approach outperforms existing methods in data-constrained scenarios, establishing a new state-of-the-art in few-shot document-level relation extraction. Our code will be publicly available.