Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AMR-RE: Abstract Meaning Representations for Retrieval-Based In-Context Learning in Relation Extraction

Peitao Han, Lis Kanashiro Pereira, Fei Cheng, Wan Jou She, Eiji Aramaki

TL;DR

This work addresses the limitation of in-context learning for relation extraction that emphasizes surface text similarity over semantic structure. It introduces AMR-RE, which parses inputs into Abstract Meaning Representation graphs, encodes them with a self-supervised SS-GNN, and retrieves demonstrations via semantic-structure similarity, focusing on the shortest AMR path between entities. In supervised experiments, AMR-RE achieves state-of-the-art results on multiple datasets and shows robust gains in unsupervised settings, outperforming sentence-embedding baselines across all evaluated datasets. The approach demonstrates that incorporating semantic graphs for demonstration retrieval improves relation extraction and offers a generalizable strategy for retrieval-based ICL in other structured prediction tasks.

Abstract

Existing in-context learning (ICL) methods for relation extraction (RE) often prioritize language similarity over structural similarity, which can lead to overlooking entity relationships. To address this, we propose an AMR-enhanced retrieval-based ICL method for RE. Our model retrieves in-context examples based on semantic structure similarity between task inputs and training samples. Evaluations on four standard English RE datasets show that our model outperforms baselines in the unsupervised setting across all datasets. In the supervised setting, it achieves state-of-the-art results on three datasets and competitive results on the fourth.

AMR-RE: Abstract Meaning Representations for Retrieval-Based In-Context Learning in Relation Extraction

TL;DR

This work addresses the limitation of in-context learning for relation extraction that emphasizes surface text similarity over semantic structure. It introduces AMR-RE, which parses inputs into Abstract Meaning Representation graphs, encodes them with a self-supervised SS-GNN, and retrieves demonstrations via semantic-structure similarity, focusing on the shortest AMR path between entities. In supervised experiments, AMR-RE achieves state-of-the-art results on multiple datasets and shows robust gains in unsupervised settings, outperforming sentence-embedding baselines across all evaluated datasets. The approach demonstrates that incorporating semantic graphs for demonstration retrieval improves relation extraction and offers a generalizable strategy for retrieval-based ICL in other structured prediction tasks.

Abstract

Existing in-context learning (ICL) methods for relation extraction (RE) often prioritize language similarity over structural similarity, which can lead to overlooking entity relationships. To address this, we propose an AMR-enhanced retrieval-based ICL method for RE. Our model retrieves in-context examples based on semantic structure similarity between task inputs and training samples. Evaluations on four standard English RE datasets show that our model outperforms baselines in the unsupervised setting across all datasets. In the supervised setting, it achieves state-of-the-art results on three datasets and competitive results on the fourth.
Paper Structure (23 sections, 2 equations, 4 figures, 5 tables)

This paper contains 23 sections, 2 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Task Definition
  4. Prompt Construction
  5. The AMR-RE Model
  6. AMR Graph Encoding
  7. Graph Representation for RE
  8. AMR-Based Demonstration Retrieval
  9. Supervised Setting
  10. Unsupervised Setting
  11. Demonstration Retrieval
  12. Experiments
  13. Main Results
  14. Results in the Supervised Setting
  15. Results in the Unsupervised Setting
  16. ...and 8 more sections

Figures (4)

  • Figure 1: An overview of our proposed method in the Supervised Setting (Section \ref{['sec:model']}, Section \ref{['sec:sup_retri']}). Given a test input, we first adopt our AMR-enhanced demonstration retrieval method to select proper demonstrations from the training set. Subsequently, all retrieved demonstrations are included in the prompt construction.
  • Figure 2: Performance for the different number of few-shot examples on TB-Dense.
  • Figure 3: A case study of semantic structure similarity. The demonstration with similar semantic structure enables the LLM to correctly generate the gold label, "Cause-Effect".
  • Figure 4: A case study of AMR-RE retrieved demonstration quality. MESSAGE AND TOPIC is the gold label.