Modeling promotion factors using Bayesian networks and video games

Abstract

The advent of electronic media, social networks and the transformation of infotainment solutions into an everyday habit has attracted the interest of creative promotion, resulting in one of the highest revenue business of today. The vast number of products and services that need to be promoted through the aforementioned channels has augmented the need for innovative, fresh and foremost successful ideas from the campaign designer perspective in order to attract customers. Towards this direction, a reasoning tool for supporting ad managers in setting up a new campaign is more than useful. This paper describes an innovative approach, in which a video game is developed in order to collect and annotate data from existing ad videos, including subjective information as regards to the impact they pose on the players. Upon collection of data, Bayesian networks are learned in order to create a mechanism for decision support and reasoning on which factors play the most important role towards establishing a fruitful promotional campaign. Due to the characteristics of the collected data, such as sparsity and high dimensionality, we propose the use of a Markov Chain Monte Carlo (MCMC) Bayesian inference engine that can cope with the above issues and provide a robust decision support tool for promotion experts. Experimental results support our claim that Bayesian networks can effectively deal with domains of high ambiguity such as the task at hand.