Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Path-based Explanation for Knowledge Graph Completion

Heng Chang, Jiangnan Ye, Alejo Lopez Avila, Jinhua Du, Jia Li

TL;DR

The proposed Power-Link is the first path-based KGC explainer that explores GNN-based models, and designs a novel simplified graph-powering technique, which enables the generation of path-based explanations with a fully parallelisable and memory-efficient training scheme.

Abstract

Graph Neural Networks (GNNs) have achieved great success in Knowledge Graph Completion (KGC) by modelling how entities and relations interact in recent years. However, the explanation of the predicted facts has not caught the necessary attention. Proper explanations for the results of GNN-based KGC models increase model transparency and help researchers develop more reliable models. Existing practices for explaining KGC tasks rely on instance/subgraph-based approaches, while in some scenarios, paths can provide more user-friendly and interpretable explanations. Nonetheless, the methods for generating path-based explanations for KGs have not been well-explored. To address this gap, we propose Power-Link, the first path-based KGC explainer that explores GNN-based models. We design a novel simplified graph-powering technique, which enables the generation of path-based explanations with a fully parallelisable and memory-efficient training scheme. We further introduce three new metrics for quantitative evaluation of the explanations, together with a qualitative human evaluation. Extensive experiments demonstrate that Power-Link outperforms the SOTA baselines in interpretability, efficiency, and scalability.

Path-based Explanation for Knowledge Graph Completion

TL;DR

The proposed Power-Link is the first path-based KGC explainer that explores GNN-based models, and designs a novel simplified graph-powering technique, which enables the generation of path-based explanations with a fully parallelisable and memory-efficient training scheme.

Abstract

Graph Neural Networks (GNNs) have achieved great success in Knowledge Graph Completion (KGC) by modelling how entities and relations interact in recent years. However, the explanation of the predicted facts has not caught the necessary attention. Proper explanations for the results of GNN-based KGC models increase model transparency and help researchers develop more reliable models. Existing practices for explaining KGC tasks rely on instance/subgraph-based approaches, while in some scenarios, paths can provide more user-friendly and interpretable explanations. Nonetheless, the methods for generating path-based explanations for KGs have not been well-explored. To address this gap, we propose Power-Link, the first path-based KGC explainer that explores GNN-based models. We design a novel simplified graph-powering technique, which enables the generation of path-based explanations with a fully parallelisable and memory-efficient training scheme. We further introduce three new metrics for quantitative evaluation of the explanations, together with a qualitative human evaluation. Extensive experiments demonstrate that Power-Link outperforms the SOTA baselines in interpretability, efficiency, and scalability.
Paper Structure (27 sections, 17 equations, 4 figures, 13 tables, 1 algorithm)

This paper contains 27 sections, 17 equations, 4 figures, 13 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Knowledge Graph Completion (KGC)
  4. GNN and KGC Explanation
  5. Paths on Graphs
  6. Preliminaries
  7. Knowledge Graphs (KGs)
  8. GNN-Based framework for KGC
  9. Problem Formulation
  10. Proposed Method: Power-Link
  11. Triplet Edge Scorer
  12. Path-Enforcing Learning
  13. Path Generation
  14. Experiments
  15. Experimental Setup
  16. ...and 12 more sections

Figures (4)

  • Figure 1: Example of the advantage that path-based explanation has over subgraph-based ones.
  • Figure 2: The overall framework of the proposed Power-Link. Given a KG and a trained KGC model as inputs, we aim to generate interpretable paths to explain why the KGC model predicts a target triplet is factual.
  • Figure 3: The explanations (green and blue arrows) by different explainers for the prediction $\langle Limitless, release\_in, Argentina \rangle$ (dashed red). Power-Link explains the fact by the $release\_in$ relationship, whereas baseline explanations are less interpretable.
  • Figure 4: The GPU memory usage of Power-Link during the explaining process against the number of edges in each graph. 500 samples are explained. For better illustration, we only record the memory usage of the explaining module. The memory occupied by the KGC model is ignored.