Monitoring Attributed Social Networks Based on Count Data and Random Effects

Abstract

This paper presents a novel approach for the statistical monitoring of online social networks where the edges represent the count of communications between ties at each time stamp. Since the available methods in the literature are limited to the assumption that the set of all interacting individuals is fixed during the monitoring horizon and their corresponding attributes do not change over time, the proposed method tackles these limitations due to the properties of the random effect concepts. Applying appropriate parameters estimation technique involved in a likelihood ratio testing (LRT) approach considering two different statistics, the longitudinal network data are monitored. The performance of the proposed method is verified using numerical examples including simulation studies as well as an illustrative example.

Zero-Truncated Poisson Regression for Zero-Inflated Multiway Count Data

“…Introduction. Count data arises in many data science applications including poll analysis [22], network communications [28,12], single photon count imaging [34,35], and ecology [5]. Statistical interpretation of count data typically involves estimating parametric distributions likely to generate the counts via regression and maximum likelihood estimation [29,14,3].…”