Quantifying Uncertainty for Temporal Motif Estimation in Graph Streams under Sampling

TL;DR

This paper develops statistical methods to quantify uncertainty in counting temporal motifs in graph streams, enabling confidence interval construction and hypothesis testing for sampled dynamic networks.

Contribution

It introduces the first asymptotic normality results for a Horvitz-Thompson estimator of temporal motif counts under sampling in deterministic and stochastic graph streams.

Findings

Proves consistency and asymptotic normality of the estimator.

Enables confidence interval construction for motif counts.

Applicable to various real-world network analysis tasks.

Abstract

Dynamic networks, a.k.a. graph streams, consist of a set of vertices and a collection of timestamped interaction events (i.e., temporal edges) between vertices. Temporal motifs are defined as classes of (small) isomorphic induced subgraphs on graph streams, considering both edge ordering and duration. As with motifs in static networks, temporal motifs are the fundamental building blocks for temporal structures in dynamic networks. Several methods have been designed to count the occurrences of temporal motifs in graph streams, with recent work focusing on estimating the count under various sampling schemes along with concentration properties. However, little attention has been given to the problem of uncertainty quantification and the asymptotic statistical properties for such count estimators. In this work, we establish the consistency and the asymptotic normality of a certain Horvitz-Thompson type of estimator in an edge sampling framework for deterministic graph streams, which can be used to construct confidence intervals and conduct hypothesis testing for the temporal motif count under sampling. We also establish similar results under an analogous stochastic model. Our results are relevant to a wide range of applications in social, communication, biological, and brain networks, for tasks involving pattern discovery.