Interpretable Multi-View Clustering

TL;DR

This work tackles the need for interpretability in multi-view clustering by proposing an interpretable MVC framework that jointly optimizes per-view embeddings and a decision-tree-based explanation. It initializes with per-view autoencoders to derive embeddings, concatenates them to form pseudo-labels $Y'$, and builds a tree in the original feature space, then iterates between refining representations (via cross-entropy guidance and soft assignments $s_{ij}^v$) and updating the tree structure. The approach achieves competitive clustering performance against state-of-the-art MVC methods while providing transparent, rule-based decision processes, as demonstrated by detailed tree visualizations. This framework opens a path toward trustworthy multi-view clustering by integrating interpretability directly into the MVC optimization loop.

Abstract

Multi-view clustering has become a significant area of research, with numerous methods proposed over the past decades to enhance clustering accuracy. However, in many real-world applications, it is crucial to demonstrate a clear decision-making process-specifically, explaining why samples are assigned to particular clusters. Consequently, there remains a notable gap in developing interpretable methods for clustering multi-view data. To fill this crucial gap, we make the first attempt towards this direction by introducing an interpretable multi-view clustering framework. Our method begins by extracting embedded features from each view and generates pseudo-labels to guide the initial construction of the decision tree. Subsequently, it iteratively optimizes the feature representation for each view along with refining the interpretable decision tree. Experimental results on real datasets demonstrate that our method not only provides a transparent clustering process for multi-view data but also delivers performance comparable to state-of-the-art multi-view clustering methods. To the best of our knowledge, this is the first effort to design an interpretable clustering framework specifically for multi-view data, opening a new avenue in this field.

Figures (6)

• Figure 1: The joint optimization framework for interpretable MVC.
• Figure 2: Illustration of objective function computation for an internal node in a decision tree. The figure shows the initial node configuration to the left of the dashed line. The right side of the dashed line demonstrates the effect of altering the splitting condition.
• Figure 3: Comparison of interpretability performance, focusing on the maximum and average depth of decision trees constructed by different interpretable clustering algorithms.
• Figure 4: The effect of parameter $\lambda$ ($y$-axis) in terms of Purity, ACC and F1-measure.
• Figure 5: The effect of parameter $maxDep$ ($y$-axis) in terms of Purity, ACC and F1-measure.