A social perspective on perceived distances reveals deep community structure

Area of Science:

• Data analysis
• Network science
• Social network analysis

Background:

• Understanding community structure is crucial for interpreting complex systems.
• Existing methods often require specific parameters or assumptions that limit their applicability.
• Data with varying density, common in dynamic processes, poses challenges for traditional analysis.

Purpose of the Study:

• To develop a new approach for capturing meaningful structural information from dissimilarity-based data.
• To introduce a measure of local community depth for probabilistic partitioning.
• To enable the identification of community structure without requiring additional inputs or assumptions.

Main Methods:

• Leveraging social concepts of conflict and alignment for local comparisons.
• Introducing a measure of local community depth.
• Developing a universal threshold for distinguishing cohesive pairs.
• Applying the method to diverse datasets including linguistics, cultural psychology, genetics, and benchmark clustering data.

Main Results:

• A novel probabilistic partitioning method is presented, conveying locally interpreted closeness (cohesion).
• The approach effectively identifies meaningful community structure across datasets with varying densities.
• It demonstrates the ability to identify structure without needing the number of clusters, neighborhood size, optimization criteria, or distributional assumptions.

Conclusions:

• The proposed method offers a robust way to identify community structure by focusing on local comparisons and cohesion.
• Its inherent recalibration to data density bypasses the need for localizing parameters common in other methods.
• This approach provides a universal framework applicable to diverse fields and data types.