AI Paper Atlas

Research map / Graph learning / GNN architectures

Reinforced Causal Explainer for Graph Neural Networks

Xiang Wang, Yingxin Wu, An Zhang, Fuli Feng, Xiangnan He, Tat‐Seng Chua

TPAMI 2022 · Graph learning › GNN architectures · approach: GNN explainability · 69 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace the strongest baseline (CXPlain) with RC-Explainertarget GIN · MutagenicityACC@10%: 0.868 -> 0.986Improved
replace the strongest baseline (PGM-Explainer) with RC-Explainertarget k-GNN · REDDIT-MULTI-5KACC@10%: 0.300 -> 0.472Improved
replace Grad-CAM with RC-Explainertarget APPNP · Visual GenomeACC@10%: 0.894 -> 0.976Improved
replace Grad-CAM with RC-Explainertarget APPNP · Visual GenomeACC-AUC: 0.917 -> 0.901Hurt
replace PGExplainer with RC-Explainertarget GIN · MutagenicityACC-AUC: 0.899 -> 0.964Improved
replace CXPlain with RC-Explainertarget k-GNN · REDDIT-MULTI-5KACC-AUC: 0.439 -> 0.521Improved
use causal screening (CXPlain, PGM-Explainer, RC-Explainer) instead of statistical attribution (SA, Grad-CAM, GNNExplainer, PGExplainer)explainers · three datasetspredictive accuracyImproved
replace PGExplainer with RC-Explainertarget GIN · Mutagenicitycontrastivity (lower better): 0.202 -> 0.311Hurt
replace CXPlain with RC-Explainertarget APPNP · Visual Genomecontrastivity (lower better): 0.320 -> 0.306Improved
replace PGM-Explainer with RC-Explainertarget k-GNN · REDDIT-MULTI-5Kcontrastivity (lower better): 0.061 -> 0.481Hurt
replace GNNExplainer with RC-Explainertarget GIN · Mutagenicitysanity check (lower better): 0.124 -> 0.248Hurt
replace GNNExplainer with RC-Explainertarget GIN · Mutagenicitytime per graph: 2.57 -> 0.680Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy