A Multi-view Clustering Model Based on Graph Contrastive Learning

Multi-view clustering, which segments nodes in multi-view networks into categories or communities, holds significant research value across various fields. However, most current single-view clustering algorithms are limited to processing single-view data, while real-world graph data are inherently mo...

Full description

Saved in:
Bibliographic Details
Published in2024 43rd Chinese Control Conference (CCC) pp. 8624 - 8630
Main Authors Wu, Bangsheng, Song, Yurong, Li, Ruqi
Format Conference Proceeding
LanguageEnglish
Published Technical Committee on Control Theory, Chinese Association of Automation 28.07.2024
Subjects
Accuracy
Aggregates
Attention Mechanism
Attention mechanisms
Clustering algorithms
Contrastive learning
Graph convolutional networks
Multi-view Clustering
Nonhomogeneous media
Self-supervised Learning
Online AccessGet full text

Cover

More Information
Summary:Multi-view clustering, which segments nodes in multi-view networks into categories or communities, holds significant research value across various fields. However, most current single-view clustering algorithms are limited to processing single-view data, while real-world graph data are inherently more complex, often requiring multi-view representations for accurate depiction. To address this, we propose a versatile and effective multi-view clustering autoencoder framework employing multilayer Graph Convolutional Networks (MVCAN) to encode multi-view networks and learn node representations. Enhanced node embedding is achieved through attention mechanisms and contrastive learning, capturing node associations more effectively. Additionally, a self-supervised module optimizes clustering centroids, and view fusion aggregates information from multiple perspectives. Experiments on ACM, DBLP, and IMDB datasets demonstrate that our MVCAN model surpasses comparative models in accuracy, confirming its effectiveness.
ISSN:1934-1768
DOI:10.23919/CCC63176.2024.10662080