Generating Intervention Hypotheses using Explainable Explanations on Graphs: G2I, a Two-Stage Greedy Framework
This work reframe counterfactual explanation as an intervention design problem, and generates counterfactuals via a greedy search that directly identifies minimal, actionable changes to node features and neighbor-level conditions that are converted into interpretable rules suitable for real-world intervention.