A Bayesian approach to out-of-sample network reconstruction

2026-02-25

AI Summary

This study addresses the challenge of accurately forecasting the future structure of dynamic networks under partial observability and scarce new data. The authors propose a self-sustaining extrapolation framework grounded in Bayesian inference, which constructs a temporal prior from historical network snapshots and integrates a single-parameter fitness model to estimate link probabilities based on node strength. Crucially, the method propagates uncertainty forward in time to enable continuous prediction without frequent retraining. Its key innovation lies in treating predicted snapshots as reliable priors, allowing iterative updates and reconstruction of subsequent network structures with minimal additional data. Empirical evaluation on interbank electronic deposit market data from 1999 to 2012 demonstrates that the approach significantly outperforms established link prediction baselines in forecasting each bank’s degree.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Machine Learning: Bayesian Learning

Application Category

• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

Networks underpin systems that range from finance to biology, yet their structure is often only partially observed. Current reconstruction methods typically fit the parameters of a model anew to each snapshot, thus offering no guidance to predict future configurations. Here, we develop a Bayesian approach that uses the information about past network snapshots to inform a prior and predict the subsequent ones, while quantifying uncertainty. Instantiated with a single-parameter fitness model, our method infers link probabilities from node strengths and carries information forward in time. When applied to the Electronic Market for Interbank Deposit across the years 1999-2012, our method accurately recovers the number of connections per bank at subsequent times, outperforming probabilistic benchmarks designed for analogous, link prediction tasks. Notably, each predicted snapshot serves as a reliable prior for the next one, thus enabling self-sustained, out-of-sample reconstruction of evolving networks with a minimal amount of additional data.

Problem

• network reconstruction
• out-of-sample prediction
• evolving networks
• partial observation
• Bayesian inference

Innovation

• Bayesian network reconstruction
• out-of-sample prediction
• temporal network dynamics
• fitness model
• link probability inference