MUSE-Explainer: Counterfactual Explanations for Symbolic Music Graph Classification Models

2025-09-30

AI Summary

Graph neural networks (GNNs) for symbolic music analysis suffer from poor interpretability, hindering their trustworthy deployment. Method: We propose a music-structure-aware counterfactual explanation framework that generates minimal, syntactically valid, and musically coherent perturbations to induce verifiable prediction changes. Our approach integrates domain-specific musical priors—including pitch, rhythm, and harmonic constraints—and employs a graph-structure-aware perturbation mechanism to ensure logical musicality. Explanations are visualized using standard tools (e.g., Verovio). Contribution/Results: Evaluated on multiple symbolic music classification tasks, our method produces intuitive, human-readable, and empirically verifiable explanations. It significantly enhances model decision transparency and establishes the first GNN explanation framework for symbolic music that jointly satisfies formal rigor and domain-specific practicality.

Technology Category

• Reasoning under Uncertainty: Graphical Models
• Computer Vision: Visual Reasoning & Symbolic Representations
• Machine Learning: Probabilistic Circuits and Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Interpretability is essential for deploying deep learning models in symbolic music analysis, yet most research emphasizes model performance over explanation. To address this, we introduce MUSE-Explainer, a new method that helps reveal how music Graph Neural Network models make decisions by providing clear, human-friendly explanations. Our approach generates counterfactual explanations by making small, meaningful changes to musical score graphs that alter a model's prediction while ensuring the results remain musically coherent. Unlike existing methods, MUSE-Explainer tailors its explanations to the structure of musical data and avoids unrealistic or confusing outputs. We evaluate our method on a music analysis task and show it offers intuitive insights that can be visualized with standard music tools such as Verovio.

Problem

• Generates counterfactual explanations for music graph classification models
• Ensures explanations remain musically coherent through meaningful graph modifications
• Provides human-interpretable insights visualizable with standard music tools

Innovation

• Generates counterfactual explanations for music graph models
• Modifies musical score graphs to alter model predictions
• Ensures explanations remain musically coherent and realistic