Subgraph Propagation and Contrastive Calibration for Incomplete Multiview Data Clustering

Area of Science:

• Data Mining
• Machine Learning
• Artificial Intelligence

Background:

• Multiview data mining is crucial but challenged by incomplete attributes due to noise and collection failures.
• Existing methods struggle with missing data, failing to calibrate complemented representations with common information across views.
• A significant issue is the cluster distribution unaligned problem (CDUP) in the latent space of incomplete multiview data.

Purpose of the Study:

• To propose a novel deep clustering framework, subgraph propagation and contrastive calibration (SPCC), for incomplete multiview raw data.
• To address the challenges of mining topology in missing multiview data and calibrating complemented representations.
• To solve the cluster distribution unaligned problem (CDUP) in the latent space of incomplete multiview data.

Main Methods:

• Reconstructing a global structural graph by propagating subgraphs from complete data in each view.
• Completing and calibrating missing views guided by the global structural graph and employing contrastive learning between views.
• Aligning complemented cluster distributions across different views using contrastive learning (CL) to resolve CDUP.

Main Results:

• The proposed SPCC framework demonstrates advanced performance on six benchmark datasets.
• The method effectively addresses the challenges of incomplete multiview data, including topology mining and representation calibration.
• Validation confirms the effectiveness and superiority of the SPCC approach in multiview clustering tasks.

Conclusions:

• The SPCC framework offers a robust solution for clustering incomplete multiview data.
• Subgraph propagation and contrastive calibration are effective strategies for handling missing data and aligning latent representations.
• The study validates the SPCC method's ability to achieve superior performance in complex multiview data mining scenarios.