Reimagining GNN Explanations with ideas from Tabular Data

TL;DR

This paper explores how explainability methods for Graph Neural Networks can be improved by drawing insights from techniques used for tabular data, focusing on Entity Matching tasks.

Contribution

It introduces a novel perspective by integrating ideas from tabular data explainability to enhance GNN explanation methods for Entity Matching.

Findings

Identifies gaps in current GNN explainability techniques.

Highlights the potential of tabular data explainability ideas for GNNs.

Proposes directions for future research in GNN explanations.

Abstract

Explainability techniques for Graph Neural Networks still have a long way to go compared to explanations available for both neural and decision decision tree-based models trained on tabular data. Using a task that straddles both graphs and tabular data, namely Entity Matching, we comment on key aspects of explainability that are missing in GNN model explanations.