Loki: A privacy-conscious platform for crowdsourced surveys.

Emerging platforms such as Amazon Mechanical Turk and Google Consumer Surveys are increasingly being used by researchers and market analysts to crowdsource large-scale survey data from on-line populations at extremely low-cost. However, by participating in successive surveys, users risk being profiled and targeted, both by surveyors and by the platform itself. In this paper we propose, develop, and evaluate the design of a crowdsourcing platform, called Loki, that is privacy conscious. Our contributions are three-fold: (a) We propose Loki, a system that allows users to obfuscate their (ratings-based or multiple-choice) responses at-source based on their chosen privacy level, and gives surveyors aggregated population averages with known statistical confidence. (b) We develop a novel selection mechanism, which the platform can use to give surveyors accurate population estimates within a cost budget, while ensuring fairness in privacy loss amongst users. (c) We evaluate our scheme both off-line using a large dataset of movie ratings, and on-line via experimentation with 131 real users using a prototype implementation on mobile devices. Our work represents a first step towards incorporating privacy protection in emerging platforms for crowdsourced survey data.