Backbone extraction through statistical edge filtering: A comparative study

Area of Science:

• Network Science
• Data Analysis
• Computational Social Science

Background:

• Backbone extraction is crucial for simplifying complex networks, aiding analysis and visualization.
• Statistical hypothesis-testing methods are widely used for filtering network edges to identify essential structures.

Purpose of the Study:

• To systematically compare seven prominent statistical hypothesis-testing backbone edge filtering methods.
• To evaluate how different methods impact network properties and edge significance.
• To provide guidance for selecting appropriate backbone extraction techniques based on network characteristics.

Main Methods:

• Systematic comparison of seven backbone extraction filters: Disparity Filter (DF), Polya Urn Filter (PF), Marginal Likelihood Filter (MLF), Noise Corrected (NC), Enhanced Configuration Model Filter (ECM), Global Statistical Significance Filter (GloSS), and Locally Adaptive Network Sparsification Filter (LANS).
• Similarity analysis of extracted backbones.
• Correlation analysis between edge features (weight, degree, betweenness) and significance levels.
• Global properties analysis (edge/node/weight fraction, entropy, reachability, components, transitivity) and distribution analysis (weight, degree).

Main Results:

• ECM and DF filters yield backbones with minimal overlap with others. A hierarchical ordering of methods (GloSS to NC, PF, LANS, MLF) shows output encapsulation.
• DF and LANS favor high-weighted edges; ECM prioritizes low significance for high-degree edges. Edge betweenness has limited impact.
• LANS preserves node count and weight entropy. DF, PF, ECM, GloSS reduce network size. MLF, NC, ECM maintain connectivity and weight entropy.
• PF and NC capture original weight distribution well. NC and MLF excel in preserving degree distribution.

Conclusions:

• Different backbone extraction methods significantly alter network properties and structural representation.
• The choice of method depends on desired outcomes, such as preserving specific edge features, network size, connectivity, or distribution patterns.
• Insights guide researchers in selecting optimal backbone extraction techniques for network analysis and visualization tasks.