Explicit View-labels Matter: A Multifacet Complementarity Study of Multi-view Clustering

TL;DR

This work reframes multi-view clustering as a multifacet problem, showing that complementarity across feature, view-label, and contrast facets substantially enhances performance beyond traditional consistency-focused MVC. It introduces MCMVC, a flexible framework that explicitly incorporates view-label information and combines reconstruction, variance, and contrast losses to learn diverse, view-specific representations without sacrificing cross-view consistency. Through extensive bi-view and multi-view experiments, MCMVC and its enhanced variant MCMVC+ consistently outperform state-of-the-art baselines across complete and incomplete MVC settings, with notable gains on Caltech101-20 and large-scale datasets like ALOI-100. The findings position view-label supervision as a strong, orthogonal signal and demonstrate the practical viability of plug-in multifacet strategies to improve clustering in heterogeneous multi-view data.

Abstract

Consistency and complementarity are two key ingredients for boosting multi-view clustering (MVC). Recently with the introduction of popular contrastive learning, the consistency learning of views has been further enhanced in MVC, leading to promising performance. However, by contrast, the complementarity has not received sufficient attention except just in the feature facet, where the Hilbert Schmidt Independence Criterion term or the independent encoder-decoder network is usually adopted to capture view-specific information. This motivates us to reconsider the complementarity learning of views comprehensively from multiple facets including the feature-, view-label- and contrast- facets, while maintaining the view consistency. We empirically find that all the facets contribute to the complementarity learning, especially the view-label facet, which is usually neglected by existing methods. Based on this, a simple yet effective \underline{M}ultifacet \underline{C}omplementarity learning framework for \underline{M}ulti-\underline{V}iew \underline{C}lustering (MCMVC) is naturally developed, which fuses multifacet complementarity information, especially explicitly embedding the view-label information. To our best knowledge, it is the first time to use view-labels explicitly to guide the complementarity learning of views. Compared with the SOTA baselines, MCMVC achieves remarkable improvements, e.g., by average margins over $5.00\%$ and $7.00\%$ respectively in complete and incomplete MVC settings on Caltech101-20 in terms of three evaluation metrics.

Figures (5)

• Figure 1: The qualitative analysis of multifacet complementarity study on Noisy MNIST.
• Figure 2: Overview of the MCMVC framework.
• Figure 3: Parameter sensitivity analysis on Caltech101-20.
• Figure 4: Clustering performance of MCMVC with increasing epoch on Caltech101-20. The x-axis denotes the training epoch, the left and right y-axis denote the clustering performance and corresponding loss value, respectively.
• Figure 5: Parameter sensitivity analysis on Caltech-5V.