Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Revisiting Document-Level Relation Extraction with Context-Guided Link Prediction

Monika Jain, Raghava Mutharaju, Ramakanth Kavuluru, Kuldeep Singh

TL;DR

This work reframes document-level relation extraction as a knowledge-graph link prediction problem augmented with external context from Wikidata and WordNet. The proposed DocRE-CLiP framework integrates a triplet-extraction step, context construction, a KG-style link predictor, and a multi-faceted reasoning module (intra-sentence, logical, and co-reference) whose outputs are aggregated for final relation predictions, with traversal-paths provided as explanations via path-based beam search. Empirical results on DocRED, ReDocRED, and DWIE show consistent gains over strong baselines, with ablations indicating context paths as a key driver of performance and explainability. The approach advances DocRE by fusing context-rich knowledge with structured reasoning, yielding improved accuracy and interpretable predictions that can bolster downstream tasks and user trust.

Abstract

Document-level relation extraction (DocRE) poses the challenge of identifying relationships between entities within a document as opposed to the traditional RE setting where a single sentence is input. Existing approaches rely on logical reasoning or contextual cues from entities. This paper reframes document-level RE as link prediction over a knowledge graph with distinct benefits: 1) Our approach combines entity context with document-derived logical reasoning, enhancing link prediction quality. 2) Predicted links between entities offer interpretability, elucidating employed reasoning. We evaluate our approach on three benchmark datasets: DocRED, ReDocRED, and DWIE. The results indicate that our proposed method outperforms the state-of-the-art models and suggests that incorporating context-based link prediction techniques can enhance the performance of document-level relation extraction models.

Revisiting Document-Level Relation Extraction with Context-Guided Link Prediction

TL;DR

This work reframes document-level relation extraction as a knowledge-graph link prediction problem augmented with external context from Wikidata and WordNet. The proposed DocRE-CLiP framework integrates a triplet-extraction step, context construction, a KG-style link predictor, and a multi-faceted reasoning module (intra-sentence, logical, and co-reference) whose outputs are aggregated for final relation predictions, with traversal-paths provided as explanations via path-based beam search. Empirical results on DocRED, ReDocRED, and DWIE show consistent gains over strong baselines, with ablations indicating context paths as a key driver of performance and explainability. The approach advances DocRE by fusing context-rich knowledge with structured reasoning, yielding improved accuracy and interpretable predictions that can bolster downstream tasks and user trust.

Abstract

Document-level relation extraction (DocRE) poses the challenge of identifying relationships between entities within a document as opposed to the traditional RE setting where a single sentence is input. Existing approaches rely on logical reasoning or contextual cues from entities. This paper reframes document-level RE as link prediction over a knowledge graph with distinct benefits: 1) Our approach combines entity context with document-derived logical reasoning, enhancing link prediction quality. 2) Predicted links between entities offer interpretability, elucidating employed reasoning. We evaluate our approach on three benchmark datasets: DocRED, ReDocRED, and DWIE. The results indicate that our proposed method outperforms the state-of-the-art models and suggests that incorporating context-based link prediction techniques can enhance the performance of document-level relation extraction models.
Paper Structure (22 sections, 7 equations, 4 figures, 5 tables)

This paper contains 22 sections, 7 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Methodology
  4. Problem Formulation
  5. Approach
  6. Triplet Extraction Module.
  7. Context Module.
  8. Link Prediction Module.
  9. Reasoning Module.
  10. Aggregation Module.
  11. Path-Based Beam Search
  12. Experimental Setup
  13. Datasets
  14. Baseline Models for Comparison
  15. Hyper-Parameters and Metrics
  16. ...and 7 more sections

Figures (4)

  • Figure 1: A partial document and labeled relation from DocRED. Blue color represents concerned entities, pink color represents other mentioned entities, and yellow color denotes the sentence number.
  • Figure 2: Triples constructed using N-hop path extracted from Wikidata. The head and tail entities are blue in color. Intermediate entities are in peach color.
  • Figure 3: Illustration of proposed framework DocRE-CLiP and its various modules.
  • Figure 4: Performance of DocRE-CLiP across various contexts using the DocRED, ReDocRED, and DWIE datasets