We present a system, predicting the point in time when users of a public display are about to leave. The ability to react to users’ intention to leave is valuable for researchers and practitioners alike: users can be presented additional content with the goal to maximize interaction times; they can be offered a discount coupon for redemption in a nearby store hence enabling new business models; or feedback can be collected from users right after they have finished interaction without interrupting their task. Our research consists of multiple steps: (1) We identified features that hint at users’ intention to leave from observations and video logs. (2) We implemented a system capable of detecting such features from Microsoft Kinect’s skeleton data and subsequently make a prediction. (3) We trained and deployed a prediction system with a Quiz game that reacts when users are about to leave (N=249), achieving an accuracy of 78%. The majority of users indeed reacted to the presented intervention.    «

We present a system, predicting the point in time when users of a public display are about to leave. The ability to react to users’ intention to leave is valuable for researchers and practitioners alike: users can be presented additional content with the goal to maximize interaction times; they can be offered a discount coupon for redemption in a nearby store hence enabling new business models; or feedback can be collected from users right after they have finished interaction without interrupt...    »