Is Meta-Path Attention an Explanation? Evidence of Alignment and Decoupling in Heterogeneous GNNs

TL;DR

The paper interrogates whether semantic-level meta-path attention in heterogeneous GNNs faithfully reflects meta-path importance. It introduces MetaXplain, a meta-path-consistent post-hoc explanation protocol, and MP-AEA, a diagnostic measuring alignment between attention and explanation-derived meta-path contributions. Across ACM, DBLP, and IMDB with HAN and HAN-GCN, lifted explanations generally outperform random baselines, but attention-importance alignment varies by dataset and backbone, revealing both alignment and decoupling regimes. Additionally, explanations can yield recoverable predictive signal when subgraphs induced by explanations are used for retraining, suggesting an explanation-as-denoising effect and highlighting nuanced implications for interpretability in heterogeneous graphs.

Abstract

Meta-path-based heterogeneous graph neural networks aggregate over meta-path-induced views, and their semantic-level attention over meta-path channels is widely used as a narrative for ``which semantics matter.'' We study this assumption empirically by asking: when does meta-path attention reflect meta-path importance, and when can it decouple? A key challenge is that most post-hoc GNN explainers are designed for homogeneous graphs, and naive adaptations to heterogeneous neighborhoods can mix semantics and confound perturbations. To enable a controlled empirical analysis, we introduce MetaXplain, a meta-path-aware post-hoc explanation protocol that applies existing explainers in the native meta-path view domain via (i) view-factorized explanations, (ii) schema-valid channel-wise perturbations, and (iii) fusion-aware attribution, without modifying the underlying predictor. We benchmark representative gradient-, perturbation-, and Shapley-style explainers on ACM, DBLP, and IMDB with HAN and HAN-GCN, comparing against xPath and type-matched random baselines under standard faithfulness metrics. To quantify attention reliability, we propose Meta-Path Attention--Explanation Alignment (MP-AEA), which measures rank correlation between learned attention weights and explanation-derived meta-path contribution scores across random runs. Our results show that meta-path-aware explanations typically outperform random controls, while MP-AEA reveals both high-alignment and statistically significant decoupling regimes depending on the dataset and backbone; moreover, retraining on explanation-induced subgraphs often preserves, and in some noisy regimes improves, predictive performance, suggesting an explanation-as-denoising effect.

Figures (7)

• Figure 1: MetaXplain lifts GNNExplainer to the meta-path view domain.
• Figure 2: Macro-F1 under controlled semantic attention interventions. Error bars show std. over 5 runs.
• Figure 3: Learned semantic attention weights. Error bars indicate std.
• Figure 4: Recoverability from explanation-induced subgraphs. Shaded regions show std. over 5 runs.
• Figure 5: Micro-F1 under controlled semantic attention interventions. Error bars show std. over 5 runs.