Multi-view graph embedding clustering network: Joint self-supervision and block diagonal representation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Multi-view clustering is a growing area in AI, but lacks methods for graph-structured data.
• Existing multi-view construction methods are limited to Euclidean data structures.

Purpose of the Study:

• To introduce a novel Multi-View Graph embedding Clustering network (MVGC) for graph-structured data.
• To address the gap in multi-view clustering for non-Euclidean data.

Main Methods:

• Leverages Euler transform to augment node attributes for non-Euclidean data.
• Imposes a block diagonal representation constraint using the ℓ_1,2-norm on the self-expression coefficient matrix.
• Integrates clustering and representation learning by using learned labels to guide model training.

Main Results:

• MVGC demonstrates superior performance compared to 11 state-of-the-art methods across four benchmark datasets.
• Achieved high accuracy on ACM (96.17%) and IMDB (53.31%) datasets.
• Showcased significant performance improvements, up to 2.85% and 1.97% respectively, over the strongest baseline.

Conclusions:

• MVGC effectively handles non-Euclidean graph-structured data for multi-view clustering.
• The proposed method achieves state-of-the-art clustering performance.
• Seamless integration of representation learning and clustering enhances overall effectiveness.