Clustering Enhanced Multiplex Graph Contrastive Representation Learning.

Area of Science:

• Graph Representation Learning
• Machine Learning
• Data Mining

Background:

• Multiplex graph representation learning methods often overlook latent community structures.
• Existing approaches fail to fully explore consistent and complementary information across different relation types.

Purpose of the Study:

• To propose a clustering-enhanced multiplex graph contrastive representation learning model (CEMR).
• To address the limitations of existing methods by discovering community structures and exploiting cross-relation information.

Main Methods:

• CEMR utilizes a multiview graph clustering framework, treating each relation type as a distinct view.
• It incorporates cross-view contrastive learning with a novel negative pair selection mechanism.
• A cross-view cosupervision module guides clustering using complementary information across views.

Main Results:

• CEMR successfully discovers potential community structures within multiplex graphs.
• The model effectively explores consistent and complementary information across different relation types.
• Experimental results demonstrate the superiority of CEMR over state-of-the-art methods on four benchmark datasets.

Conclusions:

• CEMR enhances multiplex graph representation learning by integrating community detection and cross-view learning.
• The proposed model offers a more comprehensive approach to capturing complex relationships in multiplex graphs.
• CEMR represents a significant advancement in the field, outperforming existing techniques.