MFC-ACL: Multi-view fusion clustering with attentive contrastive learning.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Multi-view clustering is crucial for high-dimensional big data mining.
• Existing methods often fail to capture holistic attribute information due to ignoring inter-view disparities.

Purpose of the Study:

• To propose an effective multi-view clustering approach that addresses limitations of current models.
• To enhance the performance of multi-view clustering by better utilizing information from multiple data views.

Main Methods:

• Developed an Attentive Autoencoder (Att-AE) module for effective view feature extraction with global information.
• Introduced a Transformer Feature Fusion Contrastive Module (TFFC) for contrastive learning of multi-view features.
• Clustered high-level features with shared consistency information for optimized results.

Main Results:

• The proposed Multi-View Fusion Clustering with Attentive Contrastive Learning (MFC-ACL) approach was evaluated.
• Experimental results demonstrated superior clustering performance compared to state-of-the-art methods.
• The approach showed effectiveness on six benchmark datasets.

Conclusions:

• MFC-ACL effectively handles multi-view data by integrating attention mechanisms and contrastive learning.
• The method successfully captures holistic attribute information and mitigates issues from view disparities.
• This work offers a significant advancement in multi-view clustering for big data applications.