Related Experiment Videos
Latent geometry of bipartite networks
Maksim Kitsak1, Fragkiskos Papadopoulos2, Dmitri Krioukov3
1Department of Physics, Northeastern University, Boston, Massachusetts 02115, USA.
Physical Review. E
|April 19, 2017
Summary
This study reveals a latent geometric structure in bipartite networks, offering a new way to analyze their organization beyond traditional projections. This approach better explains network properties and can improve systems like recommenders.
Area of Science:
- Network Science
- Complex Systems
- Data Analysis
Background:
- Bipartite networks are common but their organizing principles are understudied.
- Traditional analysis via one-mode projections loses information and creates artificial connections.
Purpose of the Study:
- To explore a latent metric structure approach for analyzing bipartite networks.
- To demonstrate this model explains real-world bipartite network properties.
- To propose a method for inferring latent distances and assess projection information loss.
Main Methods:
- Analysis of a simple latent-geometric model for bipartite networks.
- Evaluation of common neighbor distributions and bipartite clustering.
- Assessment of information loss in one-mode projections.
Main Results:
- The latent geometric model successfully explains key structural properties of bipartite networks.
- One-mode projections result in significant information loss.
- An efficient method for inferring latent pairwise distances was developed.
Conclusions:
- Bipartite networks possess an underlying latent metric structure.
- This geometric approach offers a more informative analysis than traditional projections.
- Applications include recommender systems and understanding biological networks.