Pairwise maximum-entropy model for reading fixation data

Abstract

Different people may have mixed interests and preferences when reading certain texts as well as dissimilar capacities to process information. We are interested in knowing if some types of texts induce a correlated reading behaviour among different readers. To study this problem, we run a reading experiment using high precision eye-tracking equipment that allows us to collect data of gaze times and locations for people reading diverse texts. Based on previous reading behaviour research, we expect that such information about the fixations will give us some insight about attention while reading. For the data analysis, we use a pairwise maximum-entropy inference model to examine the relations between the readers’ fixation configurations. In this approach, we interpret the correlations between these fixation patterns as “interactions" between the readers. A thermodynamic analysis shows that, for meaningful texts, the system is near a critical point, which suggests that readers correlate their attention behaviour when reading coherent texts.