Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learn to Unlearn: Meta-Learning-Based Knowledge Graph Embedding Unlearning

Naixing Xu, Qian Li, Xu Wang, Bingchen Liu, Xin Li

TL;DR

MetaEU presents a meta-learning-based framework for knowledge graph embedding unlearning, addressing privacy-driven demands to remove the influence of specific data while preserving performance on remaining data. It combines two KG-aware modules, RAEEG and NEEM, with an ensemble learning/unlearning strategy to generate high-quality obfuscated embeddings and dilute forgotten information, generalizing to unseen entities. The approach is trained via a cross-task meta-learning regime on subgraphs and evaluated on FB15k-237 across multiple KGE models, showing effective forgetting with minimal loss to test performance. The work demonstrates practical potential for privacy-preserving KG applications and highlights future directions in multi-source KG fusion and broader unlearning strategies.

Abstract

Knowledge graph (KG) embedding methods map entities and relations into continuous vector spaces, improving performance in tasks like link prediction and question answering. With rising privacy concerns, machine unlearning (MU) has emerged as a critical AI technology, enabling models to eliminate the influence of specific data. Existing MU approaches often rely on data obfuscation and adjustments to training loss but lack generalization across unlearning tasks. This paper introduces MetaEU, a Meta-Learning-Based Knowledge Graph Embedding Unlearning framework. MetaEU leverages meta-learning to unlearn specific embeddings, mitigating their impact while preserving model performance on remaining data. Experiments on benchmark datasets demonstrate its effectiveness in KG embedding unlearning.

Learn to Unlearn: Meta-Learning-Based Knowledge Graph Embedding Unlearning

TL;DR

MetaEU presents a meta-learning-based framework for knowledge graph embedding unlearning, addressing privacy-driven demands to remove the influence of specific data while preserving performance on remaining data. It combines two KG-aware modules, RAEEG and NEEM, with an ensemble learning/unlearning strategy to generate high-quality obfuscated embeddings and dilute forgotten information, generalizing to unseen entities. The approach is trained via a cross-task meta-learning regime on subgraphs and evaluated on FB15k-237 across multiple KGE models, showing effective forgetting with minimal loss to test performance. The work demonstrates practical potential for privacy-preserving KG applications and highlights future directions in multi-source KG fusion and broader unlearning strategies.

Abstract

Knowledge graph (KG) embedding methods map entities and relations into continuous vector spaces, improving performance in tasks like link prediction and question answering. With rising privacy concerns, machine unlearning (MU) has emerged as a critical AI technology, enabling models to eliminate the influence of specific data. Existing MU approaches often rely on data obfuscation and adjustments to training loss but lack generalization across unlearning tasks. This paper introduces MetaEU, a Meta-Learning-Based Knowledge Graph Embedding Unlearning framework. MetaEU leverages meta-learning to unlearn specific embeddings, mitigating their impact while preserving model performance on remaining data. Experiments on benchmark datasets demonstrate its effectiveness in KG embedding unlearning.

Paper Structure

This paper contains 21 sections, 13 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Definition
  4. Methodology
  5. The Framework of Meta-Learning
  6. The construction of the Dataset
  7. The Procedure of Meta-Learning
  8. The Extraction of Meta-Knowledge from Knowledge Graphs
  9. Relation-Aware Entity Embedding Generator
  10. Neighbor-Enhanced Embedding Modulator
  11. Ensemble Learning and Ensemble Unlearning
  12. Experiment
  13. Information about the Dataset
  14. Experimental Setup
  15. Implementation Details
  16. ...and 6 more sections

Figures (4)

  • Figure 1: An example of meta-learning-based knowledge graph embedding unlearning.
  • Figure 2: MetaEU innovatively incorporates the processes of ensemble learning and ensemble unlearning within the meta-learning framework.
  • Figure 3: Comparison of the performance of MetaEU and FedLU on the unlearning task.
  • Figure 4: Ablation studies on different components of MetaEU.