Modeling Video Viewing Styles with Probabilistic Mode Switching

Abstract

During video lectures, learners may have attentional modes such as “follow a lecturer's guide (speech and pointers),” “look ahead of spoken parts and actively check slide content,” and “roughly browse a slide.” The dynamic change of these modes is useful to characterize personal and/or temporary viewing styles. This paper presents a method to analyze video viewing styles through gaze behavioral data by using a probabilistic generative model with a latent mode variable. In our experiments, we show that the model can infer viewers' temporal mode patterns and successfully characterize task-dependent viewing situations.