Temporal Eigenvector Centrality
Temporal Eigenvector Centrality (Dynamic Eigenvector-Based Node Importance in Time-Varying Networks) · Also known as: dynamic eigenvector centrality, time-varying eigenvector centrality, TEC, temporal communicability centrality
Temporal eigenvector centrality extends the classical eigenvector centrality to networks that change over time. By accounting for the ordering and timing of connections, it identifies nodes that are influential not merely because of many simultaneous connections, but because they sit at the crossroads of sequentially important pathways across multiple time slices of the network.
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 temporal eigenvector centrality when you have a network that changes over time — such as communication logs, co-authorship streams, financial transaction records, or epidemic contact sequences — and you need to identify nodes whose influence propagates through temporally ordered chains of contact rather than through simultaneous connections. It is especially valuable when the sequence of interactions matters for spread or influence. Do not use it when the network is essentially static, when time stamps are unreliable or arbitrarily assigned, when the number of events per slice is too sparse to support meaningful matrix operations, or when a simpler temporal degree or betweenness measure already answers the research question.
Strengths & limitations
- Captures time-respecting influence: rewards nodes that sit at causally coherent pathways, not just high-degree positions.
- Sensitive to the arrow of time, making it appropriate for diffusion, epidemic, and information-flow studies.
- The supracentrality formulation naturally handles weighted, directed, and multilayer temporal networks.
- Provides per-slice centrality trajectories that reveal how node importance evolves across the study period.
- Grounded in well-established linear-algebra theory, ensuring mathematical interpretability.
- Computationally expensive for large networks with many time slices, as it requires eigendecomposition of a potentially massive supra-adjacency matrix.
- Results are sensitive to the choice of time-slice granularity and inter-layer coupling parameter, which can be difficult to justify without domain knowledge.
- Sparse slices — few edges per snapshot — produce unstable and noisy centrality estimates.
- Less intuitive to explain to non-technical audiences than degree or betweenness centrality.
Frequently asked
How is temporal eigenvector centrality different from applying eigenvector centrality to an aggregated network?
Aggregation collapses all time slices into one static graph, destroying temporal ordering. Temporal eigenvector centrality uses the ordered product of slice matrices (or a supra-adjacency matrix), so only time-respecting paths — where earlier edges causally precede later ones — contribute to a node's score. Two nodes can have the same aggregated degree but very different temporal centrality if their connections appear in different sequences.
How do I choose the right slice granularity?
Start with a granularity that matches the natural rhythm of the process (e.g., daily slices for daily communication data). Check that each slice has enough edges to support meaningful computation — extremely sparse slices should be merged. Sensitivity analysis across two or three granularity levels is strongly recommended.
What is the inter-layer coupling parameter in the supracentrality approach?
It controls how strongly a node's centrality in one time slice influences its centrality in adjacent slices. A coupling of zero treats each slice independently; a very high coupling approximates static centrality. The value should reflect how fast node importance is expected to change — slower dynamics warrant stronger coupling.
Can the method handle directed and weighted networks?
Yes. The supracentrality formulation naturally accommodates directed and weighted edges within each slice, as well as asymmetric inter-layer couplings. The ordered-product formulation also supports directed graphs by preserving edge direction in matrix multiplication.
How large a network can be handled in practice?
Exact eigendecomposition of the full supra-adjacency matrix scales poorly with many nodes and many slices. For networks with thousands of nodes and hundreds of slices, iterative methods (e.g., power iteration) or sparse approximations are necessary. Software libraries such as NetworkX combined with scipy.sparse are a common starting point.
Sources
- Grindrod, P., Parsons, M. C., Higham, D. J., & Estrada, E. (2011). Communicability across evolving networks. Physical Review E, 83(4), 046120. DOI: 10.1103/PhysRevE.83.046120 ↗
- Taylor, D., Myers, S. A., Clauset, A., Porter, M. A., & Mucha, P. J. (2017). Eigenvector-based centrality measures for temporal networks. Multiscale Modeling and Simulation, 15(1), 537-574. DOI: 10.1137/16M1066142 ↗
How to cite this page
ScholarGate. (2026, June 3). Temporal Eigenvector Centrality (Dynamic Eigenvector-Based Node Importance in Time-Varying Networks). ScholarGate. https://scholargate.app/en/network-analysis/temporal-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.
- Eigenvector CentralityNetwork analysis↔ compare
- Temporal Betweenness CentralityNetwork analysis↔ compare
- Temporal Degree CentralityNetwork analysis↔ compare
- Temporal PageRankNetwork analysis↔ compare
- Temporal Social Network AnalysisNetwork analysis↔ compare