Global-Graph Guided and Local-Graph Weighted Contrastive Learning for Unified Clustering on Incomplete and Noise Multi-View Data

TL;DR

This work tackles clustering on incomplete and noisy multi-view data by identifying rare-paired and mis-paired issues in contrastive learning and proposing a unified, imputation-free framework named Global-Local Graph-guided Contrastive Learning (GLC). It combines a global-graph guided contrastive component with a local-graph weighted contrastive component to exploit cross-view information while mitigating noise, framed within a two-stage feature learning process. Empirical results across multiple datasets under incomplete, noise, and combined perturbations demonstrate significant improvements over state-of-the-art methods, with strong ablation showing the complementary value of each module. The approach offers a practical, robust solution for real-world multi-view data analysis where missing views and noise are common, enabling reliable clustering without data imputation.

Abstract

Recently, contrastive learning (CL) plays an important role in exploring complementary information for multi-view clustering (MVC) and has attracted increasing attention. Nevertheless, real-world multi-view data suffer from data incompleteness or noise, resulting in rare-paired samples or mis-paired samples which significantly challenges the effectiveness of CL-based MVC. That is, rare-paired issue prevents MVC from extracting sufficient multi-view complementary information, and mis-paired issue causes contrastive learning to optimize the model in the wrong direction. To address these issues, we propose a unified CL-based MVC framework for enhancing clustering effectiveness on incomplete and noise multi-view data. First, to overcome the rare-paired issue, we design a global-graph guided contrastive learning, where all view samples construct a global-view affinity graph to form new sample pairs for fully exploring complementary information. Second, to mitigate the mis-paired issue, we propose a local-graph weighted contrastive learning, which leverages local neighbors to generate pair-wise weights to adaptively strength or weaken the pair-wise contrastive learning. Our method is imputation-free and can be integrated into a unified global-local graph-guided contrastive learning framework. Extensive experiments on both incomplete and noise settings of multi-view data demonstrate that our method achieves superior performance compared with state-of-the-art approaches.

Figures (6)

• Figure 1: Framework Overview of GLC. Our method consists of two stages: View-Specific Feature Learning and Global-Local Graph-Guided Contrastive Learning. In the first stage, view-specific autoencoders are trained with the reconstruction loss $\mathcal{L}_{\text{rec}}$ to extract view-specific latent features $\{\mathbf{Z}^v\}_{v=1}^{V}$, without imputing the missing data. In the second stage, we introduce (a) Global-Graph Guided Contrastive Learning with loss $\mathcal{L}_{\text{ggc}}$, where all view samples are integrated to build a global-view affinity graph for establishing new sample pairs with semantic association hidden in all views; and (b) Local-Graph Weighted Contrastive Learning with loss $\mathcal{L}_{\text{lwc}}$, which adaptively re-weights cross-view sample pairs based on local feature similarity to suppress the negative effect from noise or unreliable sample pairs.
• Figure 2: Illustration of the global-graph sample pair construction.
• Figure 3: Ablation of with/without weight $\mathcal{W}$ on four datasets.
• Figure 4: Loss $vs.$ Clustering performance on ALOI.
• Figure 5: ACC $\mathit{vs.}$ Parameters $\{\alpha,\beta\}$ on ALOI.