Hidden Markov Model and Reversible Jump Algorithm for the Analysis of Areas of Interest from Eye-Tracking Data

• Activity Number: 398
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Marketing
• Author(s): JIN YAN*+ and Michel Wedel and Paul J. Smith
• Companies: University of Maryland and University of Maryland and University of Maryland
• Keywords: Eye tracking ; Hidden Markov Model ; Reversible Jump ; Spatial Random Effect

Abstract

When looking at an advertisement, consumers' attention jumps from one point to another, until enough information is collected. By analyzing of eye tracking data from experiment, we want to find out what areas on the pages attract consumers more, and how consumers' attention jumps among them. Assuming that there are unknown number of areas of interest (AOIs) with unknown centers and sizes, and that consumers' eye-fixations are located on one of these AOIs at a time, the sequence of AOIs visited by each consumer, which is unobservable, is modeled as a Markov chain. Conditional on these hidden chains, we assume that each consumer's fixations follow a spatial Poisson distribution, whose intensity is proportional to a truncated bivariate normal density. The means of the distribution are the centers of these AOIs, and the covariance matrices represent the sizes. When the number of AOIs is fixed, the means, the covariance matrices and the transition matrices can be estimated by Markov chain Monte Carlo methods and the forward-backward algorithm. When the number of AOIs is unknown, a reversible jump Markov chain Monte Carlo algorithm is applied to estimate the number of AOIs in addition.