Research map / Graph learning / GNN architectures

The Graph Concept Bottleneck: Decoding Combinatorial Reasoning in GNNs for Interpretability

NeurIPS 2026 · Graph learning › GNN architectures · approach: GNN explainability

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes the Graph Concept Bottleneck, an interpretable framework for GNNs that exposes graph concepts with quantified contributions rather than only hard Boolean logical rules.

Problem

GNNs are black boxes, and existing methods only extract hard Boolean logical rules over graph concepts, which cannot quantify how much each concept contributes to predictions.

Approach

Combines a concept bottleneck with graph concepts so that GNN predictions are decoded into weighted contributions of topological patterns, moving beyond hard Boolean rules toward quantified combinatorial reasoning.

Experimental setup

Builds on: concept bottleneck models

Closest papers by meaning

• GraphTrail: Translating GNN Predictions into Human-Interpretable Logical Rules (NeurIPS 2024 · similarity 0.92)
• Self-Interpretable Graph Learning with Sufficient and Necessary Explanations (AAAI 2024 · similarity 0.90)
• Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · similarity 0.89)
• Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.89)
• Factorized Explainer for Graph Neural Networks (AAAI 2024 · similarity 0.89)
• Reconsidering Faithfulness in Regular, Self-Explainable and Domain Invariant GNNs (ICLR 2025 · similarity 0.89)