Gaze and visual patterns of observer while looking on visual stimuli is affected by 3 factors. First factor (bottom-up) is related to visual stimuli and its properties. Second factor (top-down) is related to personal characteristics of observer like task at hand, experience, mood or gender. Third factor is related to characteristics of oculomotor system. As a result of these factors, the observer’s gaze contains a great amount of information about the user as well as about what they are looking at.
Models, which are using information about gaze, are often based only on evaluation of aggregated metrics of gaze, thereby losing temporal information about visual exploration of observer. In this work, we focus on the creation of features by using probabilistic models, which could abstract this information from data. We investigate, which probabilistic model and with what input features has best ability to model data from selected domain and whether the subsequent clustering of created probabilistic models can improve its ability.