Auxiliary Graph for Attribute Graph Clustering.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Attribute graph clustering algorithms leverage topological information for robust node representations.
• Existing methods often overlook relationships between non-directly linked nodes, limiting clustering potential.

Purpose of the Study:

• To introduce the Auxiliary Graph for Attribute Graph Clustering (AGAGC) technique.
• To enhance attribute graph clustering by capturing broader node relationships.

Main Methods:

• Constructing an auxiliary graph based on node attributes using a noise-filtering approach.
• Training a clustering model supervised by both the original and auxiliary graphs.
• Merging multi-layer embeddings and employing a self-supervised clustering module with triplet loss.

Main Results:

• The proposed AGAGC model demonstrates competitive or superior performance compared to state-of-the-art graph clustering methods.
• Experiments conducted on four benchmark datasets validate the effectiveness of the AGAGC technique.

Conclusions:

• AGAGC effectively addresses limitations in existing attribute graph clustering by incorporating auxiliary graph information.
• The method enhances representation discriminative power and clustering awareness for improved results.