Community detection in networks using graph embeddings.

Area of Science:

• Machine Learning
• Network Science
• Data Mining

Background:

• Graph embedding techniques are increasingly utilized for node classification and link prediction.
• Embedding graphs into geometric spaces can facilitate community detection via clustering algorithms.
• The spatial proximity of nodes in embedded graphs is hypothesized to reflect network communities.

Purpose of the Study:

• To evaluate the efficacy of various graph embedding methods for community detection on benchmark graphs.
• To compare the performance of graph embedding techniques against traditional community detection algorithms.
• To assess the practicality and advantages of graph embedding for real-world network analysis.

Main Methods:

• Testing multiple graph embedding techniques on benchmark datasets.
• Comparing embedding-based community detection with established algorithms.
• Analyzing the impact of parameter selection on embedding performance.

Main Results:

• Graph embedding performance for community detection is comparable to traditional methods when parameters are optimized.
• Optimal parameter sets for graph embedding are dependent on graph characteristics, such as size.
• Traditional community detection algorithms generally do not require parameter tuning.

Conclusions:

• Current graph embedding techniques do not offer a significant improvement over network clustering algorithms for community detection.
• The need for careful, dataset-specific parameter optimization limits the practical application of graph embedding.
• High computational costs associated with embedding and clustering further reduce the advantage of these methods for community detection.