From Clustering to Cluster Explanations via Neural Networks

Explainability has been neglected in unsupervised learning: While the field of Explainable AI (XAI) has focused heavily on explaining supervised learning models like classifiers, this paper addresses a significant gap—most real-world problems involve unlabeled data where the goal is discovering clusters without explanations for why data points belong to specific groups.

A novel "neuralization" approach makes cluster explanations practical: The paper proposes converting clustering models into neural networks, which enables efficient and reliable attribution of cluster assignments to specific input features—something that "powerful methods" for clustering typically cannot do.

Cluster quality assessment becomes possible: By explaining which features drive cluster assignments, the framework allows practitioners to evaluate whether discovered clusters are meaningful and based on sensible patterns, rather than accepting clusters as a "black box" result.

The method extracts actionable insights from data: The paper demonstrates through "showcases" that explaining cluster assignments can reveal novel, interpretable patterns in data and underlying representations—suggesting the approach has practical value for understanding what structure a clustering algorithm has actually discovered.

This addresses a real gap in responsible AI practice: For organizations using clustering for tasks like customer segmentation, anomaly detection, or data exploration, the inability to explain why items cluster together creates a transparency problem that this work directly tackles.