[Paper Review] Maps of Information Flow Reveal Community Structure In Complex Networks

Martin Rosvall, Carl T. Bergstrom|arXiv (Cornell University)|Jul 4, 2007

Complex Network Analysis Techniques5 references146 citations

TL;DR

This paper introduces an information-theoretic method to uncover community structure in weighted, directed networks by optimally compressing information flows. Applied to 6,000+ journals' citation patterns, it reveals a multicentric scientific network with bidirectional flows along core fields and a directional citation flow from applied to basic sciences, highlighting the multipartite, dynamic, and integrated nature of knowledge systems.

ABSTRACT

To comprehend the multipartite organization of large-scale biological and social systems, we introduce a new information theoretic approach to reveal community structure in weighted and directed networks. The method decomposes a network into modules by optimally compressing a description of information flows on the network. The result is a map that both simplifies and highlights the regularities in the structure and their relationships. We illustrate the method by making a map of scientific communication as captured in the citation patterns of more than 6000 journals. We discover a multicentric organization with fields that vary dramatically in size and degree of integration into the network of science. Along the backbone of the network — including physics, chemistry, molecular biology, and medicine — information flows bidirectionally, but the map reveals a directional pattern of citation from the applied fields to the basic sciences. Biological and social systems are differentiated, multipartite, integrated, and dynamic. Data about these systems, now available on unprecedented scales, are often schematized as networks. Such abstractions are powerful (1, 2), but even as abstractions they remain highly complex. It is therefore helpful to decompose the myriad nodes and links into modules that

Motivation & Objective

• To address the challenge of understanding the multipartite, dynamic, and integrated organization of large-scale biological and social systems.
• To develop a method that simplifies complex network abstractions by revealing structural regularities in information flow.
• To decompose weighted and directed networks into meaningful modules based on optimal compression of flow descriptions.
• To map scientific communication patterns across journals to uncover structural and directional patterns in knowledge dissemination.

Proposed method

• The method uses information theory to model information flows across nodes in a network, treating flows as probabilistic signals.
• It applies optimal compression principles to identify modular groupings that minimize the description length of observed flows.
• The approach decomposes the network into modules by minimizing the Kullback-Leibler divergence between observed and modeled flow distributions.
• The resulting map highlights structural regularities by emphasizing high-probability, low-description-length flow patterns.
• The method is applied to citation data from over 6,000 journals to infer community structure in the network of science.
• Directionality and integration levels are inferred from the asymmetry and density of compressed flow patterns between modules.

Experimental results

Research questions

• RQ1 How can community structure in complex, weighted, and directed networks be revealed through information flow patterns?
• RQ2 What structural regularities emerge when information flows in scientific citation networks are optimally compressed?
• RQ3 How do the sizes and integration levels of scientific fields vary across the network of science?
• RQ4 What directional patterns of knowledge flow exist between applied and basic scientific fields?
• RQ5 To what extent is the network of science organized as a multicentric, multipartite system?

Key findings

• The network of science exhibits a multicentric structure with fields varying widely in size and integration levels.
• Core fields such as physics, chemistry, molecular biology, and medicine form a backbone with bidirectional information flow.
• A consistent directional pattern of citation is observed from applied fields toward basic sciences, indicating a unidirectional flow of influence.
• The method successfully reveals structural regularities by simplifying complex network abstractions into interpretable modules.
• Biological and social systems are shown to be dynamically integrated, multipartite, and differentiated in their information flow organization.
• The decomposition highlights both modular cohesion and inter-field connectivity, emphasizing the role of information compression in network analysis.