Heterogeneous Graph Neural Networks (HGNNs) have become essential for modeling complex systems with diverse, interacting components. Meta relations, which represent distinct interaction types between source and target nodes, are crucial for HGNNs to capture rich, context-specific semantic information. However, existing GNN’s explanatory models struggle to effectively capture the unique semantic aspects of HGNNs, often leading to incomplete or misleading explanations. Moreover, current explanatory models tend to assess edge significance globally, overlooking finer-grained differences among edges with shared semantic content, thus limiting their ability to provide context-specific insights. To address these limitations, we propose MR-Explainer, a Meta Relation-Assisted Explanatory Model for HGNNs. MR-Explainer incorporates a heterogeneous information fusion module to integrate structural and semantic data, generating comprehensive edge representations. Additionally, a multiplex salience-aware module computes salience values both globally and within each meta relation, ensuring that explanations are context-sensitive and precise. Evaluations on classification tasks demonstrate MR-Explainer’s effectiveness in delivering accurate, nuanced explanations for HGNN’s outcomes.