CrowdSignals

Researchers from diverse backgrounds critically depend on mobile datasets. From training and testing activity recognition models, to verifying hypotheses in social science, mobile data is indispensable. Unfortunately, mobile data collection requires significant time and budget for infrastructure as well as subject recruiting, screening, training, legal agreements, equipment, and compensation. We estimate up to 70% of the resources in a study may be spent on data collection. Moreover, this massive investment can combine with institutional, legal, and political issues to create a disincentive to sharing with the community. In this paper, we propose and justify a crowdfunded and crowdsourced methodology for longitudinal mobile data collection that cuts researcher costs by orders of magnitude, removes barriers to data sharing, and boosts data value for all stakeholders. We also present CrowdSignals, a first instantiation which will generate the largest labeled mobile dataset available to the community.