Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

InteractiveGNNExplainer: A Visual Analytics Framework for Multi-Faceted Understanding and Probing of Graph Neural Network Predictions

TC Singh, Sougata Mukherjea

TL;DR

The paper addresses the opacity of Graph Neural Networks for node classification and proposes InteractiveGNNExplainer, a visual analytics framework that fuses multi-view explanations (GNNExplainer, GAT attention) with interactive graph editing for what-if analysis. It integrates a Dash-based frontend with PyTorch Geometric back-end to allow dynamic exploration of graphs, embeddings, features, and neighborhood context, providing real-time re-inference and explanation updates. Case studies on Cora and CiteSeer demonstrate diagnosing misclassifications, comparing GCN vs GAT explanations, and probing model sensitivity. The work contributes a novel, end-to-end platform that enhances transparency, trust, and robustness in graph-based AI and offers directions for scalability and broader applicability.

Abstract

Graph Neural Networks (GNNs) excel in graph-based learning tasks, but their complex, non-linear operations often render them as opaque "black boxes". This opacity hinders user trust, complicates debugging, bias detection, and adoption in critical domains requiring explainability. This paper introduces InteractiveGNNExplainer, a visual analytics framework to enhance GNN explainability, focusing on node classification. Our system uniquely integrates coordinated interactive views (dynamic graph layouts, embedding projections, feature inspection, neighborhood analysis) with established post-hoc (GNNExplainer) and intrinsic (GAT attention) explanation techniques. Crucially, it incorporates interactive graph editing, allowing users to perform a "what-if" analysis by perturbing graph structures and observing immediate impacts on GNN predictions and explanations. We detail the system architecture and, through case studies on Cora and CiteSeer datasets, demonstrate how InteractiveGNNExplainer facilitates in-depth misclassification diagnosis, comparative analysis of GCN versus GAT behaviors, and rigorous probing of model sensitivity. These capabilities foster a deeper, multifaceted understanding of GNN predictions, contributing to more transparent, trustworthy, and robust graph analysis.

InteractiveGNNExplainer: A Visual Analytics Framework for Multi-Faceted Understanding and Probing of Graph Neural Network Predictions

TL;DR

The paper addresses the opacity of Graph Neural Networks for node classification and proposes InteractiveGNNExplainer, a visual analytics framework that fuses multi-view explanations (GNNExplainer, GAT attention) with interactive graph editing for what-if analysis. It integrates a Dash-based frontend with PyTorch Geometric back-end to allow dynamic exploration of graphs, embeddings, features, and neighborhood context, providing real-time re-inference and explanation updates. Case studies on Cora and CiteSeer demonstrate diagnosing misclassifications, comparing GCN vs GAT explanations, and probing model sensitivity. The work contributes a novel, end-to-end platform that enhances transparency, trust, and robustness in graph-based AI and offers directions for scalability and broader applicability.

Abstract

Graph Neural Networks (GNNs) excel in graph-based learning tasks, but their complex, non-linear operations often render them as opaque "black boxes". This opacity hinders user trust, complicates debugging, bias detection, and adoption in critical domains requiring explainability. This paper introduces InteractiveGNNExplainer, a visual analytics framework to enhance GNN explainability, focusing on node classification. Our system uniquely integrates coordinated interactive views (dynamic graph layouts, embedding projections, feature inspection, neighborhood analysis) with established post-hoc (GNNExplainer) and intrinsic (GAT attention) explanation techniques. Crucially, it incorporates interactive graph editing, allowing users to perform a "what-if" analysis by perturbing graph structures and observing immediate impacts on GNN predictions and explanations. We detail the system architecture and, through case studies on Cora and CiteSeer datasets, demonstrate how InteractiveGNNExplainer facilitates in-depth misclassification diagnosis, comparative analysis of GCN versus GAT behaviors, and rigorous probing of model sensitivity. These capabilities foster a deeper, multifaceted understanding of GNN predictions, contributing to more transparent, trustworthy, and robust graph analysis.

Paper Structure

This paper contains 25 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. GNN Explanation Methods
  4. Perturbation-based Methods
  5. Gradient-based and Feature Attribution Methods
  6. Surrogate Models
  7. Attention-based Explanations
  8. Counterfactual and Factual Reasoning
  9. Visual Analytics for Deep Learning
  10. InteractiveGNNExplainer System
  11. Design Goals and Principles
  12. System Architecture
  13. Backend (Python)
  14. Frontend (Dash Application)
  15. Model Training Details
  16. ...and 10 more sections

Figures (5)

  • Figure 1: High-level architecture: User interacts with UI; Dash Callbacks trigger Backend Modules (Data, Model, Inference, Explanation, Editing) which update UI views.
  • Figure 2: Overview of the InteractiveGNNExplainer dashboard: (a) Control Panel, (b) Graph View, (c) Embeddings View, (d) Feature Importance, (e) Selected Info, (f) Neighbor Analysis, (g) Explanation View.
  • Figure 3: Initial analysis of misclassified Node 1536 on Cora. Graph view shows GNNExplainer edge highlights. Neighbor analysis indicates peer misclassifications.
  • Figure 4: Node 1536 after removing an influential edge. Prediction changes correctly to theory(True) value, reflecting model sensitivity.
  • Figure 5: GAT explanation for a node on CiteSeer: GNNExplainer subgraph (top) and GAT attention scores (bottom) illustrate focused importance from the attention mechanism.