Deep Multi-View Clustering via Cluster-Semantic Guidance

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Deep multi-view clustering leverages heterogeneous data for relationship discovery.
• Existing models struggle with inter-cluster separability and semantic integration, limiting performance.

Purpose of the Study:

• To propose a novel deep multi-view clustering method using cluster-semantic guidance.
• To enhance feature discriminability and clustering performance by addressing limitations of current models.

Main Methods:

• Separating clusters to improve inter-cluster discriminability.
• Employing knowledge distillation for cluster stability and representation learning.
• Aggregating sample-level semantic information for a cluster-oriented learning strategy.

Main Results:

• The method learns discriminative and clustering-friendly representations.
• It strengthens sample representation capability through a cluster-oriented perspective.
• Experiments show superior clustering performance on various datasets.

Conclusions:

• The proposed method effectively enhances deep multi-view clustering.
• It achieves state-of-the-art performance by learning distinctive feature embeddings.