EDGE: Evaluation of Diverse Knowledge Graph Explanations

Researcher

Rupesh Sapkota

Publications

• Rupesh Sapkota, Dominik Köhler, Stefan Heindorf (2024): EDGE: Evaluation Framework for Logical vs. Subgraph Explanations for Node Classifiers on Knowledge Graphs Full Publication

Research Theme

R3 Sustainability & Efficiency

Tags

Class Expression Learning Explainable AI Knowledge Graphs Neural Networks

The project focuses on an evaluation framework called EDGE to evaluate and compare different types of explanations for Graph Neural Networks (GNNs) used in knowledge graph node classification. The aim is to assess the effectiveness of explanations such as logical rule-based explanations and subgraph-based explanations in terms of prediction accuracy and fidelity to the GNN models. This evaluation is crucial to improve the transparency and interpretability of AI systems, especially as GNNs are often treated as “black-box” models.