Improving Human Answers Quality by Machine Questions Number and Context Factors

Mobile phones provide an opportunity for a symbiotic interaction between humans and machines, which allows phones to collect human-centric data at anytime and anywhere. However, low-quality answers, which refer to the wrong answers, may be provided by users when they are asked excessive questions or in unsuitable contexts (e.g., driving). To solve this problem, we aim to design a methodology to collect more correct answers. We propose to use answer reaction time to annotate answer quality, to find a suitable number of daily questions, and the context factors that need to be considered according to their history records. We validated our methodology via the public dataset, which was collected by an extensive four-week in-the-wild study at the University of Trento, Italy. The results reveal that the context information and the number of daily questions are factors that can impact user answer behavior. These factors, therefore, influence the answer quality.