HiSE: A Lightweight Hierarchical Semantic Explainer for Heterogeneous Graph Neural Networks

2026-06-02

AI Summary

This work addresses the limited interpretability of heterogeneous graph neural networks (HGNNs) in high-stakes scenarios, where existing explanation methods struggle to capture semantic hierarchies and suffer from high computational costs. To overcome these challenges, we propose HiSE—a lightweight, feature-oriented hierarchical semantic explanation framework. HiSE constructs sparse local surrogate models at the semantic level using LASSO and adaptively fuses multi-view contributions across semantic layers via KL divergence. By introducing a hierarchical semantic modeling mechanism for the first time, HiSE significantly enhances explanation fidelity and cross-semantic coherence while drastically reducing computational overhead. Extensive experiments demonstrate that HiSE consistently outperforms state-of-the-art baselines in terms of fidelity, robustness, and scalability, making it well-suited for large-scale real-world heterogeneous graphs.

Abstract

Heterogeneous graph neural networks (HGNNs) have demonstrated remarkable performance in modeling complex relational data, however their interpretability in high-stakes applications remains a critical challenge. Existing explanation methods suffer from two major limitations: on the one hand, the generated explanations fail to reflect the inherent semantic hierarchy of HGNNs, resulting in a lack of fidelity to the model's internal decision-making mechanism; on the other hand, feature explanations often rely on complex search or perturbation mechanisms, leading to excessive computational complexity and poor efficiency. To address these issues, we propose HiSE, a lightweight feature-oriented interpretable model for HGNNs. HiSE achieves semantically aware feature explanations through hierarchical semantic modeling: at the semantic level, local surrogate models based on the Least Absolute Shrinkage and Selection Operator (LASSO) are employed to learn sparse feature representations under each semantic view; at the cross-semantic level, the contributions of different semantic views are adaptively characterized via KL divergence to produce a unified explanation. Extensive experiments demonstrate that HiSE outperforms existing methods in terms of fidelity, robustness, and cross-semantic explanation capability, while its lightweight framework incurs low computational overhead, enabling efficient application to large-scale, complex real-world heterogeneous graphs.

Problem

• heterogeneous graph neural networks
• interpretability
• semantic hierarchy
• computational complexity
• feature explanation

Innovation

• hierarchical semantic explanation
• heterogeneous graph neural networks
• LASSO-based surrogate model
• KL divergence
• lightweight interpretability