Dynamic Eigenvector Centrality
Dynamic Eigenvector Centrality in Temporal Networks · Also known as: temporal eigenvector centrality, time-varying eigenvector centrality, dynamic EC, evolving eigenvector centrality
Dynamic eigenvector centrality extends the classic eigenvector centrality measure to networks that change over time. Rather than computing a single leading eigenvector on a static adjacency matrix, it tracks how a node's influence — defined by the importance of its neighbours — evolves across snapshots or time windows. The method is used in social network analysis, epidemiology, and information diffusion studies where network topology shifts continuously.
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 dynamic eigenvector centrality when your network has meaningful temporal variation and the research question concerns how node influence evolves over time — for example, identifying persistently influential actors in communication networks, tracking the rise and fall of influential users on social media, or studying how infection spreads through a contact network that changes daily. It is not appropriate when the network is effectively static (edge turnover is negligible), when snapshots are so sparse that adjacency matrices are disconnected and eigenvectors become unreliable, or when a single cross-sectional centrality ranking fully answers the research question. Avoid it when the network has very few nodes per snapshot or when edge weights are not meaningful across time periods.
Strengths & limitations
- Captures the time-evolving nature of influence in networks where static measures would miss transient hubs.
- Anchored in the well-understood eigenvector centrality framework, making results intuitive to network analysts.
- Flexible aggregation strategies (mean, decay-weighted, cumulative) allow adaptation to different research questions.
- Produces longitudinal centrality trajectories that support change-point detection and trend analysis.
- Compatible with weighted and directed network snapshots when the underlying eigenvector method is extended accordingly.
- Snapshot granularity is a subjective choice that can strongly affect results; no universal rule governs window width.
- Sparse snapshots — common when data are collected at fine time resolution — can yield disconnected graphs with poorly defined or degenerate eigenvectors.
- Computationally more expensive than static eigenvector centrality because the eigen-decomposition must be repeated for every snapshot.
- Aggregation of per-snapshot scores can obscure short-lived but important bursts of influence.
- Requires consistent node-set definition across snapshots; nodes entering or leaving the network complicate interpretation.
Frequently asked
How should I choose the snapshot window length?
There is no universal rule. Conduct a sensitivity analysis by repeating the analysis at two or three window widths (e.g., daily, weekly, monthly) and checking whether the ranking of top nodes is stable. If results change dramatically, report results under multiple granularities and discuss the substantive meaning of each.
What if some snapshots are disconnected or have very few edges?
Disconnected snapshots produce multiple eigenvectors and no unique leading centrality score. Options include merging sparse snapshots into wider windows, using the largest connected component only (noting the limitation), or applying a regularisation term (e.g., adding a small constant to the diagonal) before computing the eigenvector.
How is dynamic eigenvector centrality different from temporal PageRank?
Both extend eigenvector-style influence to temporal networks. PageRank adds a damping factor and handles directed links naturally. Dynamic eigenvector centrality is the more direct generalisation of Bonacich centrality, and may be preferable in undirected or symmetric contact networks where the PageRank teleportation assumption is harder to justify.
Can I use this with weighted edges?
Yes. Replace binary adjacency matrices with weighted adjacency matrices in each snapshot. Weighted eigenvector centrality (via the dominant eigenvector of the weighted matrix) then reflects not only who a node connects to but how strongly, making it sensitive to tie strength variation across time.
Is there software support for dynamic eigenvector centrality?
General network libraries such as NetworkX (Python) and igraph (R/Python) compute eigenvector centrality per snapshot. Dedicated temporal network packages (e.g., teneto in Python) provide higher-level wrappers for dynamic centrality calculations including aggregation across snapshots.
Sources
How to cite this page
ScholarGate. (2026, June 3). Dynamic Eigenvector Centrality in Temporal Networks. ScholarGate. https://scholargate.app/en/network-analysis/dynamic-eigenvector-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.
- Dynamic PageRankNetwork analysis↔ compare
- Eigenvector CentralityNetwork analysis↔ compare
- Temporal Community DetectionNetwork analysis↔ compare
- Temporal Network AnalysisNetwork analysis↔ compare