Bayesian Betweenness Centrality
Bayesian Betweenness Centrality (Probabilistic Inference of Shortest-Path Centrality) · Also known as: Bayesian BC, probabilistic betweenness centrality, uncertainty-aware betweenness centrality, posterior betweenness estimation
Bayesian Betweenness Centrality estimates how often a node lies on shortest paths in a network while explicitly quantifying uncertainty arising from incomplete, sampled, or noisy edge observations. Rather than producing a single point estimate, it yields a posterior distribution over betweenness scores, enabling credible intervals and probabilistic comparisons between nodes.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Bayesian Betweenness Centrality when the network is derived from sampled, self-reported, or otherwise uncertain relational data — such as survey-based friendship networks, citation graphs with incomplete coverage, or communication logs that capture only a fraction of true interactions. It is appropriate when decisions hinge on the relative ranking of nodes and confidence in those rankings matters, or when the network is small enough that estimation uncertainty is non-trivial. Avoid it when the network is directly and completely observed (e.g., a full organisational email log), when computational resources are limited (MCMC inference on large graphs is expensive), or when a straightforward node ranking is sufficient without uncertainty bounds.
Strengths & limitations
- Produces credible intervals around betweenness scores, making uncertainty in rankings explicit and interpretable.
- Naturally handles incomplete or sampled network data by treating missing edges as latent variables.
- Enables probabilistic comparisons between nodes, such as the probability that node A is more central than node B.
- Incorporates domain knowledge through informative priors, which is especially valuable in small networks.
- Consistent Bayesian framework allows hierarchical extensions and integration with other probabilistic network models.
- Computationally expensive: MCMC sampling combined with repeated betweenness computation limits scalability to large networks.
- Prior specification requires careful thought; poorly chosen priors can dominate results in small samples.
- Implementation complexity is high compared to classical betweenness; fewer off-the-shelf tools are available.
- Interpreting and communicating posterior distributions to non-specialist audiences requires additional effort.
- For large, completely observed networks, the added complexity rarely yields benefits over point estimates.
Frequently asked
How does this differ from simply bootstrapping classical betweenness centrality?
Bootstrapping resamples observed edges but does not model uncertainty about whether unobserved edges exist. Bayesian betweenness centrality places an explicit probabilistic model over the true network structure, including edges that may be missing from the data, producing a principled posterior rather than an empirical sampling distribution.
What software can I use to implement Bayesian betweenness centrality?
There is no single dedicated package. A typical workflow combines a Bayesian network-sampling library (e.g., graph-tool in Python for latent network inference, or custom MCMC in Stan/PyMC) with NetworkX or igraph to compute classical betweenness on each sampled graph. The posterior is then assembled from the per-sample scores.
How large can the network be before the method becomes impractical?
Scalability depends on both the MCMC mixing time and the cost of betweenness computation (O(VE) per sample). In practice, networks with thousands of nodes and edges become computationally demanding; networks with tens of thousands of nodes typically require approximation strategies such as betweenness estimation on sampled paths or variational inference.
Is a prior over edges or over betweenness scores used directly?
The prior is placed over the network structure or edge parameters, not directly over betweenness scores. Betweenness scores are then a derived quantity computed from each sampled network, so their posterior distribution is an induced consequence of the structural prior and likelihood.
When should I report credible intervals versus a single point estimate?
Report credible intervals whenever the ranking or selection of nodes is a substantive decision — for instance, choosing intervention targets in public health or identifying key brokers in an organisation. When the network is large and well-observed and the posterior variance is negligible, a point estimate with a brief note on uncertainty is sufficient.
Sources
- Newman, M.E.J. (2010). Networks: An Introduction. Oxford University Press. ISBN: 978-0-19-920665-0
- Fortunato, S., Bergstrom, C.T., Borner, K., Evans, J.A., Helbing, D., Milojevi, S., Petersen, A.M., Radicchi, F., Sinatra, R., Uzzi, B., Vespignani, A., Waltman, L., Wang, D. & Barabasi, A.-L. (2018). Science of science. Science, 359(6379), eaao0185. DOI: 10.1126/science.aao0185 ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Betweenness Centrality (Probabilistic Inference of Shortest-Path Centrality). ScholarGate. https://scholargate.app/en/network-analysis/bayesian-betweenness-centrality
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bayesian Social Network AnalysisNetwork analysis↔ compare
- Betweenness CentralityNetwork analysis↔ compare
- Weighted Betweenness CentralityNetwork analysis↔ compare