Broad spectrum structure discovery in large-scale higher-order networks.

Area of Science:

• Network Science
• Computational Biology
• Sociology

Background:

• Complex systems exhibit higher-order interactions, often modeled as hypergraphs.
• Understanding hypergraph structures is vital but computationally challenging due to combinatorial complexity.

Purpose of the Study:

• To develop a probabilistic model for efficient representation and discovery of mesoscale structures in large-scale hypergraphs.
• To enable tractable capture of rich structural patterns and ensure model identifiability.

Main Methods:

• Introduced a novel class of probabilistic models.
• Treated classes of similar units as nodes in a latent hypergraph.
• Modeled observed interactions via latent class interactions using low-rank representations.

Main Results:

• The model efficiently represents and discovers mesoscale structures in hypergraphs.
• Achieved improved link prediction performance compared to state-of-the-art methods.
• Discovered interpretable node- and class-level structures in real-world networks.

Conclusions:

• The proposed model offers a tractable approach to analyzing large-scale hypergraphs.
• Enhances the ability to incorporate higher-order interaction data into scientific analysis.
• Demonstrates utility in diverse fields like pharmacology and social network analysis.