The article introduces MANTRA, a framework for approximating the temporal betweenness centrality in large temporal graphs. The key highlights and insights are:
MANTRA extends the state-of-the-art estimator for temporal betweenness centrality to cover different temporal path optimality criteria (shortest, shortest-foremost, prefix-foremost).
The article derives new bounds on the sufficient number of samples needed to approximate the temporal betweenness centrality for all nodes. These bounds are governed by three key quantities of the temporal graph: the temporal vertex diameter, average temporal path length, and the maximum variance of the temporal betweenness centrality estimators.
The authors propose a novel algorithm to efficiently estimate the key quantities (temporal diameter, average temporal path length, temporal connectivity rate) that the sample complexity bounds depend on. This algorithm uses a mixed approach of sampling and counting, with theoretical guarantees on the approximation quality.
The MANTRA framework incorporates the sample complexity bounds and a progressive sampling technique to provide a high-quality approximation of the temporal betweenness centrality. MANTRA improves upon the state-of-the-art ONBRA algorithm in terms of running time, sample size, and required space while maintaining high accuracy.
The article provides an extensive experimental evaluation comparing MANTRA with ONBRA on several real-world temporal networks, demonstrating the superior performance of MANTRA.
Naar een andere taal
vanuit de broninhoud
arxiv.org
Belangrijkste Inzichten Gedestilleerd Uit
by Antonio Cruc... om arxiv.org 04-09-2024
https://arxiv.org/pdf/2304.08356.pdfDiepere vragen