Detection of predictability ratings of live events on TV by use of second screens
Kiraz Candan Herdem · 2014
Event predictability, one dimension of human emotion description, indicates to what extent sequences of events in videos are predictable for viewers. This study adopts second screening style of Social TV viewing model, where viewers text about live events on TV via social media app on mobile second screen. 14 instances from different TV content types are presented to 19 viewers on TV-like screen. While texting, custom Twitter application collects touch and inertial sensors' data of which features are extracted to model viewers' physical interaction with mobile screens. Viewers self-reported their predictability ratings via a slider with 9 scales, which later are divided equally into three levels indicating whether viewers describe events in videos as unpredictable, medium or predictable. Bayesian networking classifier is created to recognize the three predictability labels from features described in physical interaction model. The study result shows that the predictability labels are recognized with 85.7% average accuracy.