Finding Counterfactual Evidences for Node Classification

2025-05-16

AI Summary

This paper addresses the dual bottlenecks of fairness and interpretability in GNN-based node classification by formally defining and solving the *node-level counterfactual evidence search* problem: identifying pairs of nodes with highly similar features and local subgraph structures yet assigned different predicted classes. To this end, we propose a model-agnostic, general-purpose search framework featuring three key innovations: (1) a hybrid similarity metric combining MLP-based feature alignment with subgraph matching for joint feature–structure comparison; (2) a pruning-enhanced bidirectional search algorithm; and (3) a structure-aware hashing index for efficient retrieval. Extensive experiments on multiple benchmark graphs demonstrate that our method significantly improves GNN fairness—reducing average statistical parity difference (SPD) by 37%—while marginally boosting classification accuracy (up to +2.1%). The generated counterfactual evidence enables effective bias diagnosis and supports robust, fairness-aware model training.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Search and Optimization: Learning to Search
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Counterfactual learning is emerging as an important paradigm, rooted in causality, which promises to alleviate common issues of graph neural networks (GNNs), such as fairness and interpretability. However, as in many real-world application domains where conducting randomized controlled trials is impractical, one has to rely on available observational (factual) data to detect counterfactuals. In this paper, we introduce and tackle the problem of searching for counterfactual evidences for the GNN-based node classification task. A counterfactual evidence is a pair of nodes such that, regardless they exhibit great similarity both in the features and in their neighborhood subgraph structures, they are classified differently by the GNN. We develop effective and efficient search algorithms and a novel indexing solution that leverages both node features and structural information to identify counterfactual evidences, and generalizes beyond any specific GNN. Through various downstream applications, we demonstrate the potential of counterfactual evidences to enhance fairness and accuracy of GNNs.

Problem

• Finding counterfactual evidences for GNN node classification
• Detecting similar nodes with different GNN classifications
• Enhancing GNN fairness and accuracy via counterfactual analysis

Innovation

• Counterfactual learning for GNN node classification
• Search algorithms for counterfactual evidence identification
• Novel indexing using features and structural information