Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Prompt-based Zero-shot Relation Extraction with Semantic Knowledge Augmentation

Jiaying Gong, Hoda Eldardiry

TL;DR

This work tackles zero-shot relation triplet extraction by introducing ZS-SKA, a prompt-based model augmented with semantic knowledge from external graphs. It combines a word-level data augmentation rule to generate unseen-relations data and knowledge-graph–driven prompts that construct weighted virtual label words, enabling semantic transfer from seen to unseen relations. The model uses a BERT-CNN instance encoder, a NER component, and prototypical networks to form relation prototypes and classify queries via Euclidean distance, with a threshold to determine unseen relations. Experiments on FewRel, Wiki-ZSL, and NYT demonstrate that ZS-SKA consistently outperforms state-of-the-art zero-shot RTE and RC baselines, especially in challenging, unbalanced settings, while showing robustness under varying unseen-relations fractions. The results highlight the practical value of integrating semantic knowledge through prompts and augmented data for open-world relation extraction tasks.

Abstract

In relation triplet extraction (RTE), recognizing unseen relations for which there are no training instances is a challenging task. Efforts have been made to recognize unseen relations based on question-answering models or relation descriptions. However, these approaches miss the semantic information about connections between seen and unseen relations. In this paper, We propose a prompt-based model with semantic knowledge augmentation (ZS-SKA) to recognize unseen relations under the zero-shot setting. We present a new word-level analogy-based sentence translation rule and generate augmented instances with unseen relations from instances with seen relations using that new rule. We design prompts with weighted virtual label construction based on an external knowledge graph to integrate semantic knowledge information learned from seen relations. Instead of using the actual label sets in the prompt template, we construct weighted virtual label words. We learn the representations of both seen and unseen relations with augmented instances and prompts. We then calculate the distance between the generated representations using prototypical networks to predict unseen relations. Extensive experiments conducted on three public datasets FewRel, Wiki-ZSL, and NYT, show that ZS-SKA outperforms other methods under zero-shot setting. Results also demonstrate the effectiveness and robustness of ZS-SKA.

Prompt-based Zero-shot Relation Extraction with Semantic Knowledge Augmentation

TL;DR

This work tackles zero-shot relation triplet extraction by introducing ZS-SKA, a prompt-based model augmented with semantic knowledge from external graphs. It combines a word-level data augmentation rule to generate unseen-relations data and knowledge-graph–driven prompts that construct weighted virtual label words, enabling semantic transfer from seen to unseen relations. The model uses a BERT-CNN instance encoder, a NER component, and prototypical networks to form relation prototypes and classify queries via Euclidean distance, with a threshold to determine unseen relations. Experiments on FewRel, Wiki-ZSL, and NYT demonstrate that ZS-SKA consistently outperforms state-of-the-art zero-shot RTE and RC baselines, especially in challenging, unbalanced settings, while showing robustness under varying unseen-relations fractions. The results highlight the practical value of integrating semantic knowledge through prompts and augmented data for open-world relation extraction tasks.

Abstract

In relation triplet extraction (RTE), recognizing unseen relations for which there are no training instances is a challenging task. Efforts have been made to recognize unseen relations based on question-answering models or relation descriptions. However, these approaches miss the semantic information about connections between seen and unseen relations. In this paper, We propose a prompt-based model with semantic knowledge augmentation (ZS-SKA) to recognize unseen relations under the zero-shot setting. We present a new word-level analogy-based sentence translation rule and generate augmented instances with unseen relations from instances with seen relations using that new rule. We design prompts with weighted virtual label construction based on an external knowledge graph to integrate semantic knowledge information learned from seen relations. Instead of using the actual label sets in the prompt template, we construct weighted virtual label words. We learn the representations of both seen and unseen relations with augmented instances and prompts. We then calculate the distance between the generated representations using prototypical networks to predict unseen relations. Extensive experiments conducted on three public datasets FewRel, Wiki-ZSL, and NYT, show that ZS-SKA outperforms other methods under zero-shot setting. Results also demonstrate the effectiveness and robustness of ZS-SKA.
Paper Structure (41 sections, 9 equations, 5 figures, 7 tables, 2 algorithms)

This paper contains 41 sections, 9 equations, 5 figures, 7 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Prompt Learning in NLP
  4. Zero-shot Relation Classification
  5. Zero-shot Relation Triplet Extraction
  6. Methodology
  7. Problem Definition
  8. Semantic Knowledge Augmentation
  9. Data Augmentation
  10. Prompts from Knowledge Graph
  11. Model Architecture and Training
  12. Instance Encoder
  13. Name Entity Extractor
  14. Model Training
  15. Experiments
  16. ...and 26 more sections

Figures (5)

  • Figure 1: Zero-shot RTE. There is no overlap of classes between training and testing data.
  • Figure 2: ZS-SKA overall architecture with components explained in Sec. \ref{['sec:3.2']}.
  • Figure 3: Denoising in virtual label construction.
  • Figure 4: Example of using Name Entity Extractor to extract relation triplets.
  • Figure 5: Effects on varying threshold $\tau$ and number of virtual labels $n$ on three datasets.