[Paper Review] Reconstructing Networks

Tiziano Squartini, Rossana Mastrandrea+1 more|arXiv (Cornell University)|Aug 31, 2021

Complex Network Analysis Techniques252 references3 citations

TL;DR

This paper presents a comprehensive overview of network reconstruction methods rooted in statistical physics and information theory, focusing on inferring missing or hidden network structures across macroscopic, mesoscopic, and microscopic scales. It introduces inference techniques to recover network topology from incomplete or noisy data, with key contributions in unifying theoretical frameworks for reconstructing complex systems from limited observations.

ABSTRACT

Complex networks datasets often come with the problem of missing information: interactions data that have not been measured or discovered, may be affected by errors, or are simply hidden because of privacy issues. This Element provides an overview of the ideas, methods and techniques to deal with this problem and that together define the field of network reconstruction. Given the extent of the subject, we shall focus on the inference methods rooted in statistical physics and information theory. The discussion will be organized according to the different scales of the reconstruction task, that is, whether the goal is to reconstruct the macroscopic structure of the network, to infer its mesoscale properties, or to predict the individual microscopic connections.

Motivation & Objective

• To address the pervasive issue of missing, erroneous, or hidden interactions in complex network datasets.
• To develop and unify inference techniques for reconstructing network structure across different scales: macroscopic, mesoscopic, and microscopic.
• To focus on methods grounded in statistical physics and information theory for principled network reconstruction.
• To provide a systematic overview of reconstruction techniques applicable to real-world complex systems with incomplete observations.

Proposed method

• Utilizes statistical inference frameworks based on maximum entropy principles to reconstruct network structures from partial data.
• Applies information-theoretic measures such as mutual information and entropy to quantify uncertainty and guide reconstruction.
• Employs probabilistic modeling to infer unobserved links while preserving observed macroscopic properties like degree distribution or clustering.
• Integrates Bayesian inference to update beliefs about network structure based on observed data and prior knowledge.
• Classifies reconstruction tasks by scale: macroscopic (global topology), mesoscopic (community structure), and microscopic (individual edges).
• Leverages constraints derived from empirical observations to constrain the space of possible network configurations.

Experimental results

Research questions

• RQ1 How can network structure be reliably inferred when interaction data is incomplete or partially observed?
• RQ2 What statistical and information-theoretic principles enable robust reconstruction of complex networks?
• RQ3 How do reconstruction methods differ in performance and applicability across macroscopic, mesoscopic, and microscopic scales?
• RQ4 What are the fundamental trade-offs between accuracy, data requirements, and computational complexity in network reconstruction?

Key findings

• The paper establishes that maximum entropy models provide a principled foundation for reconstructing networks under incomplete data constraints.
• Information-theoretic measures enable quantification of uncertainty and guide the selection of the most likely network configurations.
• Reconstruction at the macroscopic level focuses on preserving global properties such as degree sequences and clustering coefficients.
• Mesoscale reconstruction aims to recover community structures using modularity and block models under partial observation.
• Microscopic reconstruction involves predicting individual missing or unobserved edges using local connectivity patterns and probabilistic inference.
• The integration of statistical physics and information theory enables a unified framework applicable across diverse network types and data regimes.