Large-Scale Clustering With Structured Optimal Bipartite Graph.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Large-scale data clustering necessitates scalable algorithms.
• Existing bipartite graph methods and spectral embedding ignore explicit cluster structure learning.
• Current anchor-based approaches lack performance stability.

Purpose of the Study:

• To investigate scalability, stability, and integration in large-scale graph clustering.
• To propose a cluster-structured graph learning model for improved clustering.
• To develop an initialization-independent anchor selection strategy.

Main Methods:

• A novel cluster-structured graph learning model is proposed.
• The model generates a c-connected bipartite graph directly from data.
• An initialization-independent anchor selection strategy is designed.

Main Results:

• The proposed method achieves direct discrete label acquisition.
• Experimental results on synthetic and real-world datasets show superior performance.
• The model demonstrates enhanced scalability and stability.

Conclusions:

• The cluster-structured graph learning model effectively addresses limitations of existing methods.
• The approach offers a stable and scalable solution for large-scale graph clustering.
• Direct cluster label acquisition simplifies the clustering pipeline.