Multi-View Clustering With Hybrid-Order Similarity Learning

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Multi-view clustering (MVC) leverages diverse data features but often struggles to integrate first-order and topological information.
• Existing MVC methods may not fully capture complex data structures due to limitations in modeling relationships across views.

Purpose of the Study:

• To propose a novel multi-view clustering method, MCHL, that jointly models first-order and topological relationships.
• To enhance clustering by integrating view-specific graphs and learning consensus representations within a unified framework.

Main Methods:

• Developed MCHL, a multi-view clustering approach incorporating hybrid-order similarity learning.
• Integrated multiple view-specific graphs, considering their first-order and topological correlations.
• Iteratively learned view weights and a consensus graph, imposing a connectivity constraint on the consensus graph.

Main Results:

• MCHL demonstrated superior clustering performance compared to state-of-the-art methods.
• Experiments on benchmark datasets validated the effectiveness of the proposed hybrid-order similarity learning approach.
• The connectivity constraint ensured proper intra-cluster data point connections within the consensus graph.

Conclusions:

• MCHL offers a robust framework for multi-view clustering by effectively integrating heterogeneous data features.
• The method's ability to model hybrid-order similarities significantly improves clustering accuracy.
• MCHL represents a significant advancement in leveraging topological and first-order relationships for comprehensive clustering structures.