Statistically Modeling and Predicting Task Success

The PARADISE approach postulates that user satisfaction is maximized when task success rates are maximized and dialog costs minimized. While this assumption may hold for the evaluation of dialog systems and completed dialogs, task success cannot be taken into account when estimating user satisfaction during an interaction as the task success is not ascertainable when a dialog is not finished. The previously presented IQ approach is thus entirely based on efficiency and qualitative measures and omits any information (i.e. features) about task success.