DiVA-DocRE: A Discriminative and Voice-Aware Paradigm for Document-Level Relation Extraction

2024-09-07

arXiv.org

AI Summary

Existing document-level relation triple extraction (DocRTE) methods largely inherit sentence-level paradigms, struggling to model cross-sentence semantics and complex syntactic structures; moreover, their heavy reliance on prompt engineering leads to low efficiency and poor generalization. This paper proposes a two-stage discriminative framework: first classifying relation types, then jointly localizing subject and object entities—enabling end-to-end triple generation. We pioneer the formulation of DocRTE as a fine-grained discriminative task and explicitly incorporate a voice-aware module to model how active/passive voice affects relational expression, thereby enhancing semantic matching robustness. Leveraging large language models, we design a lightweight discriminative architecture that eliminates redundant prompt engineering. Our approach achieves state-of-the-art performance on both Re-DocRED and DocRED, significantly outperforming existing generative and joint extraction methods.

Technology Category

• Natural Language Processing: Information Extraction
• Reasoning under Uncertainty: Relational Probabilistic Models
• Knowledge Representation and Reasoning: Description Logics

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

The remarkable capabilities of Large Language Models (LLMs) in text comprehension and generation have revolutionized Information Extraction (IE). One such advancement is in Document-level Relation Triplet Extraction (DocRTE), a critical task in information systems that aims to extract entities and their semantic relationships from documents. However, existing methods are primarily designed for Sentence level Relation Triplet Extraction (SentRTE), which typically handles a limited set of relations and triplet facts within a single sentence. Additionally, some approaches treat relations as candidate choices integrated into prompt templates, resulting in inefficient processing and suboptimal performance when determining the relation elements in triplets. To address these limitations, we introduce a Discriminative and Voice Aware Paradigm DiVA. DiVA involves only two steps: performing document-level relation extraction (DocRE) and then identifying the subject object entities based on the relation. No additional processing is required simply input the document to directly obtain the triplets. This streamlined process more accurately reflects real-world scenarios for triplet extraction. Our innovation lies in transforming DocRE into a discriminative task, where the model pays attention to each relation and to the often overlooked issue of active vs. passive voice within the triplet. Our experiments on the Re-DocRED and DocRED datasets demonstrate state-of-the-art results for the DocRTE task.

Problem

• Addresses limitations in document-level relation triplet extraction.
• Improves efficiency by reducing steps in relation extraction process.
• Focuses on active vs passive voice discrimination in triplets.

Innovation

• Two-step discriminative DocRE and entity identification
• Focus on active vs passive voice in triplets
• Direct triplet extraction from document input