Simultaneous analysis for audience profiling and audience prediction

Abstract

TV platform operators have demographic insight of the primary account holder of a household but know little about the other residents. Operators and advertisers have to access the complete set of demographic profiles, but also estimate the demographic composition for each subscriber household to provide better services. This paper describes
a methodology for audience profiling and audience predictions. Simultaneous Analysis (SA) of contingency tables is applied to the aggregated contingency tables of users viewing event counts for multiple guidebooks; logistic regression is used on the SA factor space to estimate the household compositions for single-membership households; multimembership household demographic composition is estimated by a generative model of multiple distributions of single-membership household profiles; audience prediction is estimated as a weighted sum of class percentage of single and multi-users groups